1 /*
2  * Copyright (c) 2016-2017 Hisilicon Limited.
3  *
4  * This software is available to you under a choice of one of two
5  * licenses.  You may choose to be licensed under the terms of the GNU
6  * General Public License (GPL) Version 2, available from the file
7  * COPYING in the main directory of this source tree, or the
8  * OpenIB.org BSD license below:
9  *
10  *     Redistribution and use in source and binary forms, with or
11  *     without modification, are permitted provided that the following
12  *     conditions are met:
13  *
14  *      - Redistributions of source code must retain the above
15  *        copyright notice, this list of conditions and the following
16  *        disclaimer.
17  *
18  *      - Redistributions in binary form must reproduce the above
19  *        copyright notice, this list of conditions and the following
20  *        disclaimer in the documentation and/or other materials
21  *        provided with the distribution.
22  *
23  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
24  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
25  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
26  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
27  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
28  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
29  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
30  * SOFTWARE.
31  */
32 
33 #include <linux/acpi.h>
34 #include <linux/etherdevice.h>
35 #include <linux/interrupt.h>
36 #include <linux/iopoll.h>
37 #include <linux/kernel.h>
38 #include <linux/types.h>
39 #include <net/addrconf.h>
40 #include <rdma/ib_addr.h>
41 #include <rdma/ib_cache.h>
42 #include <rdma/ib_umem.h>
43 #include <rdma/uverbs_ioctl.h>
44 
45 #include "hnae3.h"
46 #include "hns_roce_common.h"
47 #include "hns_roce_device.h"
48 #include "hns_roce_cmd.h"
49 #include "hns_roce_hem.h"
50 #include "hns_roce_hw_v2.h"
51 
52 enum {
53 	CMD_RST_PRC_OTHERS,
54 	CMD_RST_PRC_SUCCESS,
55 	CMD_RST_PRC_EBUSY,
56 };
57 
58 enum ecc_resource_type {
59 	ECC_RESOURCE_QPC,
60 	ECC_RESOURCE_CQC,
61 	ECC_RESOURCE_MPT,
62 	ECC_RESOURCE_SRQC,
63 	ECC_RESOURCE_GMV,
64 	ECC_RESOURCE_QPC_TIMER,
65 	ECC_RESOURCE_CQC_TIMER,
66 	ECC_RESOURCE_SCCC,
67 	ECC_RESOURCE_COUNT,
68 };
69 
70 static const struct {
71 	const char *name;
72 	u8 read_bt0_op;
73 	u8 write_bt0_op;
74 } fmea_ram_res[] = {
75 	{ "ECC_RESOURCE_QPC",
76 	  HNS_ROCE_CMD_READ_QPC_BT0, HNS_ROCE_CMD_WRITE_QPC_BT0 },
77 	{ "ECC_RESOURCE_CQC",
78 	  HNS_ROCE_CMD_READ_CQC_BT0, HNS_ROCE_CMD_WRITE_CQC_BT0 },
79 	{ "ECC_RESOURCE_MPT",
80 	  HNS_ROCE_CMD_READ_MPT_BT0, HNS_ROCE_CMD_WRITE_MPT_BT0 },
81 	{ "ECC_RESOURCE_SRQC",
82 	  HNS_ROCE_CMD_READ_SRQC_BT0, HNS_ROCE_CMD_WRITE_SRQC_BT0 },
83 	/* ECC_RESOURCE_GMV is handled by cmdq, not mailbox */
84 	{ "ECC_RESOURCE_GMV",
85 	  0, 0 },
86 	{ "ECC_RESOURCE_QPC_TIMER",
87 	  HNS_ROCE_CMD_READ_QPC_TIMER_BT0, HNS_ROCE_CMD_WRITE_QPC_TIMER_BT0 },
88 	{ "ECC_RESOURCE_CQC_TIMER",
89 	  HNS_ROCE_CMD_READ_CQC_TIMER_BT0, HNS_ROCE_CMD_WRITE_CQC_TIMER_BT0 },
90 	{ "ECC_RESOURCE_SCCC",
91 	  HNS_ROCE_CMD_READ_SCCC_BT0, HNS_ROCE_CMD_WRITE_SCCC_BT0 },
92 };
93 
94 static inline void set_data_seg_v2(struct hns_roce_v2_wqe_data_seg *dseg,
95 				   struct ib_sge *sg)
96 {
97 	dseg->lkey = cpu_to_le32(sg->lkey);
98 	dseg->addr = cpu_to_le64(sg->addr);
99 	dseg->len  = cpu_to_le32(sg->length);
100 }
101 
102 /*
103  * mapped-value = 1 + real-value
104  * The hns wr opcode real value is start from 0, In order to distinguish between
105  * initialized and uninitialized map values, we plus 1 to the actual value when
106  * defining the mapping, so that the validity can be identified by checking the
107  * mapped value is greater than 0.
108  */
109 #define HR_OPC_MAP(ib_key, hr_key) \
110 		[IB_WR_ ## ib_key] = 1 + HNS_ROCE_V2_WQE_OP_ ## hr_key
111 
112 static const u32 hns_roce_op_code[] = {
113 	HR_OPC_MAP(RDMA_WRITE,			RDMA_WRITE),
114 	HR_OPC_MAP(RDMA_WRITE_WITH_IMM,		RDMA_WRITE_WITH_IMM),
115 	HR_OPC_MAP(SEND,			SEND),
116 	HR_OPC_MAP(SEND_WITH_IMM,		SEND_WITH_IMM),
117 	HR_OPC_MAP(RDMA_READ,			RDMA_READ),
118 	HR_OPC_MAP(ATOMIC_CMP_AND_SWP,		ATOM_CMP_AND_SWAP),
119 	HR_OPC_MAP(ATOMIC_FETCH_AND_ADD,	ATOM_FETCH_AND_ADD),
120 	HR_OPC_MAP(SEND_WITH_INV,		SEND_WITH_INV),
121 	HR_OPC_MAP(MASKED_ATOMIC_CMP_AND_SWP,	ATOM_MSK_CMP_AND_SWAP),
122 	HR_OPC_MAP(MASKED_ATOMIC_FETCH_AND_ADD,	ATOM_MSK_FETCH_AND_ADD),
123 	HR_OPC_MAP(REG_MR,			FAST_REG_PMR),
124 };
125 
126 static u32 to_hr_opcode(u32 ib_opcode)
127 {
128 	if (ib_opcode >= ARRAY_SIZE(hns_roce_op_code))
129 		return HNS_ROCE_V2_WQE_OP_MASK;
130 
131 	return hns_roce_op_code[ib_opcode] ? hns_roce_op_code[ib_opcode] - 1 :
132 					     HNS_ROCE_V2_WQE_OP_MASK;
133 }
134 
135 static void set_frmr_seg(struct hns_roce_v2_rc_send_wqe *rc_sq_wqe,
136 			 const struct ib_reg_wr *wr)
137 {
138 	struct hns_roce_wqe_frmr_seg *fseg =
139 		(void *)rc_sq_wqe + sizeof(struct hns_roce_v2_rc_send_wqe);
140 	struct hns_roce_mr *mr = to_hr_mr(wr->mr);
141 	u64 pbl_ba;
142 
143 	/* use ib_access_flags */
144 	hr_reg_write_bool(fseg, FRMR_BIND_EN, wr->access & IB_ACCESS_MW_BIND);
145 	hr_reg_write_bool(fseg, FRMR_ATOMIC,
146 			  wr->access & IB_ACCESS_REMOTE_ATOMIC);
147 	hr_reg_write_bool(fseg, FRMR_RR, wr->access & IB_ACCESS_REMOTE_READ);
148 	hr_reg_write_bool(fseg, FRMR_RW, wr->access & IB_ACCESS_REMOTE_WRITE);
149 	hr_reg_write_bool(fseg, FRMR_LW, wr->access & IB_ACCESS_LOCAL_WRITE);
150 
151 	/* Data structure reuse may lead to confusion */
152 	pbl_ba = mr->pbl_mtr.hem_cfg.root_ba;
153 	rc_sq_wqe->msg_len = cpu_to_le32(lower_32_bits(pbl_ba));
154 	rc_sq_wqe->inv_key = cpu_to_le32(upper_32_bits(pbl_ba));
155 
156 	rc_sq_wqe->byte_16 = cpu_to_le32(wr->mr->length & 0xffffffff);
157 	rc_sq_wqe->byte_20 = cpu_to_le32(wr->mr->length >> 32);
158 	rc_sq_wqe->rkey = cpu_to_le32(wr->key);
159 	rc_sq_wqe->va = cpu_to_le64(wr->mr->iova);
160 
161 	hr_reg_write(fseg, FRMR_PBL_SIZE, mr->npages);
162 	hr_reg_write(fseg, FRMR_PBL_BUF_PG_SZ,
163 		     to_hr_hw_page_shift(mr->pbl_mtr.hem_cfg.buf_pg_shift));
164 	hr_reg_clear(fseg, FRMR_BLK_MODE);
165 }
166 
167 static void set_atomic_seg(const struct ib_send_wr *wr,
168 			   struct hns_roce_v2_rc_send_wqe *rc_sq_wqe,
169 			   unsigned int valid_num_sge)
170 {
171 	struct hns_roce_v2_wqe_data_seg *dseg =
172 		(void *)rc_sq_wqe + sizeof(struct hns_roce_v2_rc_send_wqe);
173 	struct hns_roce_wqe_atomic_seg *aseg =
174 		(void *)dseg + sizeof(struct hns_roce_v2_wqe_data_seg);
175 
176 	set_data_seg_v2(dseg, wr->sg_list);
177 
178 	if (wr->opcode == IB_WR_ATOMIC_CMP_AND_SWP) {
179 		aseg->fetchadd_swap_data = cpu_to_le64(atomic_wr(wr)->swap);
180 		aseg->cmp_data = cpu_to_le64(atomic_wr(wr)->compare_add);
181 	} else {
182 		aseg->fetchadd_swap_data =
183 			cpu_to_le64(atomic_wr(wr)->compare_add);
184 		aseg->cmp_data = 0;
185 	}
186 
187 	hr_reg_write(rc_sq_wqe, RC_SEND_WQE_SGE_NUM, valid_num_sge);
188 }
189 
190 static int fill_ext_sge_inl_data(struct hns_roce_qp *qp,
191 				 const struct ib_send_wr *wr,
192 				 unsigned int *sge_idx, u32 msg_len)
193 {
194 	struct ib_device *ibdev = &(to_hr_dev(qp->ibqp.device))->ib_dev;
195 	unsigned int left_len_in_pg;
196 	unsigned int idx = *sge_idx;
197 	unsigned int i = 0;
198 	unsigned int len;
199 	void *addr;
200 	void *dseg;
201 
202 	if (msg_len > qp->sq.ext_sge_cnt * HNS_ROCE_SGE_SIZE) {
203 		ibdev_err(ibdev,
204 			  "no enough extended sge space for inline data.\n");
205 		return -EINVAL;
206 	}
207 
208 	dseg = hns_roce_get_extend_sge(qp, idx & (qp->sge.sge_cnt - 1));
209 	left_len_in_pg = hr_hw_page_align((uintptr_t)dseg) - (uintptr_t)dseg;
210 	len = wr->sg_list[0].length;
211 	addr = (void *)(unsigned long)(wr->sg_list[0].addr);
212 
213 	/* When copying data to extended sge space, the left length in page may
214 	 * not long enough for current user's sge. So the data should be
215 	 * splited into several parts, one in the first page, and the others in
216 	 * the subsequent pages.
217 	 */
218 	while (1) {
219 		if (len <= left_len_in_pg) {
220 			memcpy(dseg, addr, len);
221 
222 			idx += len / HNS_ROCE_SGE_SIZE;
223 
224 			i++;
225 			if (i >= wr->num_sge)
226 				break;
227 
228 			left_len_in_pg -= len;
229 			len = wr->sg_list[i].length;
230 			addr = (void *)(unsigned long)(wr->sg_list[i].addr);
231 			dseg += len;
232 		} else {
233 			memcpy(dseg, addr, left_len_in_pg);
234 
235 			len -= left_len_in_pg;
236 			addr += left_len_in_pg;
237 			idx += left_len_in_pg / HNS_ROCE_SGE_SIZE;
238 			dseg = hns_roce_get_extend_sge(qp,
239 						idx & (qp->sge.sge_cnt - 1));
240 			left_len_in_pg = 1 << HNS_HW_PAGE_SHIFT;
241 		}
242 	}
243 
244 	*sge_idx = idx;
245 
246 	return 0;
247 }
248 
249 static void set_extend_sge(struct hns_roce_qp *qp, struct ib_sge *sge,
250 			   unsigned int *sge_ind, unsigned int cnt)
251 {
252 	struct hns_roce_v2_wqe_data_seg *dseg;
253 	unsigned int idx = *sge_ind;
254 
255 	while (cnt > 0) {
256 		dseg = hns_roce_get_extend_sge(qp, idx & (qp->sge.sge_cnt - 1));
257 		if (likely(sge->length)) {
258 			set_data_seg_v2(dseg, sge);
259 			idx++;
260 			cnt--;
261 		}
262 		sge++;
263 	}
264 
265 	*sge_ind = idx;
266 }
267 
268 static bool check_inl_data_len(struct hns_roce_qp *qp, unsigned int len)
269 {
270 	struct hns_roce_dev *hr_dev = to_hr_dev(qp->ibqp.device);
271 	int mtu = ib_mtu_enum_to_int(qp->path_mtu);
272 
273 	if (mtu < 0 || len > qp->max_inline_data || len > mtu) {
274 		ibdev_err(&hr_dev->ib_dev,
275 			  "invalid length of data, data len = %u, max inline len = %u, path mtu = %d.\n",
276 			  len, qp->max_inline_data, mtu);
277 		return false;
278 	}
279 
280 	return true;
281 }
282 
283 static int set_rc_inl(struct hns_roce_qp *qp, const struct ib_send_wr *wr,
284 		      struct hns_roce_v2_rc_send_wqe *rc_sq_wqe,
285 		      unsigned int *sge_idx)
286 {
287 	struct hns_roce_dev *hr_dev = to_hr_dev(qp->ibqp.device);
288 	u32 msg_len = le32_to_cpu(rc_sq_wqe->msg_len);
289 	struct ib_device *ibdev = &hr_dev->ib_dev;
290 	unsigned int curr_idx = *sge_idx;
291 	void *dseg = rc_sq_wqe;
292 	unsigned int i;
293 	int ret;
294 
295 	if (unlikely(wr->opcode == IB_WR_RDMA_READ)) {
296 		ibdev_err(ibdev, "invalid inline parameters!\n");
297 		return -EINVAL;
298 	}
299 
300 	if (!check_inl_data_len(qp, msg_len))
301 		return -EINVAL;
302 
303 	dseg += sizeof(struct hns_roce_v2_rc_send_wqe);
304 
305 	if (msg_len <= HNS_ROCE_V2_MAX_RC_INL_INN_SZ) {
306 		hr_reg_clear(rc_sq_wqe, RC_SEND_WQE_INL_TYPE);
307 
308 		for (i = 0; i < wr->num_sge; i++) {
309 			memcpy(dseg, ((void *)wr->sg_list[i].addr),
310 			       wr->sg_list[i].length);
311 			dseg += wr->sg_list[i].length;
312 		}
313 	} else {
314 		hr_reg_enable(rc_sq_wqe, RC_SEND_WQE_INL_TYPE);
315 
316 		ret = fill_ext_sge_inl_data(qp, wr, &curr_idx, msg_len);
317 		if (ret)
318 			return ret;
319 
320 		hr_reg_write(rc_sq_wqe, RC_SEND_WQE_SGE_NUM, curr_idx - *sge_idx);
321 	}
322 
323 	*sge_idx = curr_idx;
324 
325 	return 0;
326 }
327 
328 static int set_rwqe_data_seg(struct ib_qp *ibqp, const struct ib_send_wr *wr,
329 			     struct hns_roce_v2_rc_send_wqe *rc_sq_wqe,
330 			     unsigned int *sge_ind,
331 			     unsigned int valid_num_sge)
332 {
333 	struct hns_roce_v2_wqe_data_seg *dseg =
334 		(void *)rc_sq_wqe + sizeof(struct hns_roce_v2_rc_send_wqe);
335 	struct hns_roce_qp *qp = to_hr_qp(ibqp);
336 	int j = 0;
337 	int i;
338 
339 	hr_reg_write(rc_sq_wqe, RC_SEND_WQE_MSG_START_SGE_IDX,
340 		     (*sge_ind) & (qp->sge.sge_cnt - 1));
341 
342 	hr_reg_write(rc_sq_wqe, RC_SEND_WQE_INLINE,
343 		     !!(wr->send_flags & IB_SEND_INLINE));
344 	if (wr->send_flags & IB_SEND_INLINE)
345 		return set_rc_inl(qp, wr, rc_sq_wqe, sge_ind);
346 
347 	if (valid_num_sge <= HNS_ROCE_SGE_IN_WQE) {
348 		for (i = 0; i < wr->num_sge; i++) {
349 			if (likely(wr->sg_list[i].length)) {
350 				set_data_seg_v2(dseg, wr->sg_list + i);
351 				dseg++;
352 			}
353 		}
354 	} else {
355 		for (i = 0; i < wr->num_sge && j < HNS_ROCE_SGE_IN_WQE; i++) {
356 			if (likely(wr->sg_list[i].length)) {
357 				set_data_seg_v2(dseg, wr->sg_list + i);
358 				dseg++;
359 				j++;
360 			}
361 		}
362 
363 		set_extend_sge(qp, wr->sg_list + i, sge_ind,
364 			       valid_num_sge - HNS_ROCE_SGE_IN_WQE);
365 	}
366 
367 	hr_reg_write(rc_sq_wqe, RC_SEND_WQE_SGE_NUM, valid_num_sge);
368 
369 	return 0;
370 }
371 
372 static int check_send_valid(struct hns_roce_dev *hr_dev,
373 			    struct hns_roce_qp *hr_qp)
374 {
375 	struct ib_device *ibdev = &hr_dev->ib_dev;
376 
377 	if (unlikely(hr_qp->state == IB_QPS_RESET ||
378 		     hr_qp->state == IB_QPS_INIT ||
379 		     hr_qp->state == IB_QPS_RTR)) {
380 		ibdev_err(ibdev, "failed to post WQE, QP state %u!\n",
381 			  hr_qp->state);
382 		return -EINVAL;
383 	} else if (unlikely(hr_dev->state >= HNS_ROCE_DEVICE_STATE_RST_DOWN)) {
384 		ibdev_err(ibdev, "failed to post WQE, dev state %d!\n",
385 			  hr_dev->state);
386 		return -EIO;
387 	}
388 
389 	return 0;
390 }
391 
392 static unsigned int calc_wr_sge_num(const struct ib_send_wr *wr,
393 				    unsigned int *sge_len)
394 {
395 	unsigned int valid_num = 0;
396 	unsigned int len = 0;
397 	int i;
398 
399 	for (i = 0; i < wr->num_sge; i++) {
400 		if (likely(wr->sg_list[i].length)) {
401 			len += wr->sg_list[i].length;
402 			valid_num++;
403 		}
404 	}
405 
406 	*sge_len = len;
407 	return valid_num;
408 }
409 
410 static __le32 get_immtdata(const struct ib_send_wr *wr)
411 {
412 	switch (wr->opcode) {
413 	case IB_WR_SEND_WITH_IMM:
414 	case IB_WR_RDMA_WRITE_WITH_IMM:
415 		return cpu_to_le32(be32_to_cpu(wr->ex.imm_data));
416 	default:
417 		return 0;
418 	}
419 }
420 
421 static int set_ud_opcode(struct hns_roce_v2_ud_send_wqe *ud_sq_wqe,
422 			 const struct ib_send_wr *wr)
423 {
424 	u32 ib_op = wr->opcode;
425 
426 	if (ib_op != IB_WR_SEND && ib_op != IB_WR_SEND_WITH_IMM)
427 		return -EINVAL;
428 
429 	ud_sq_wqe->immtdata = get_immtdata(wr);
430 
431 	hr_reg_write(ud_sq_wqe, UD_SEND_WQE_OPCODE, to_hr_opcode(ib_op));
432 
433 	return 0;
434 }
435 
436 static int fill_ud_av(struct hns_roce_v2_ud_send_wqe *ud_sq_wqe,
437 		      struct hns_roce_ah *ah)
438 {
439 	struct ib_device *ib_dev = ah->ibah.device;
440 	struct hns_roce_dev *hr_dev = to_hr_dev(ib_dev);
441 
442 	hr_reg_write(ud_sq_wqe, UD_SEND_WQE_UDPSPN, ah->av.udp_sport);
443 	hr_reg_write(ud_sq_wqe, UD_SEND_WQE_HOPLIMIT, ah->av.hop_limit);
444 	hr_reg_write(ud_sq_wqe, UD_SEND_WQE_TCLASS, ah->av.tclass);
445 	hr_reg_write(ud_sq_wqe, UD_SEND_WQE_FLOW_LABEL, ah->av.flowlabel);
446 
447 	if (WARN_ON(ah->av.sl > MAX_SERVICE_LEVEL))
448 		return -EINVAL;
449 
450 	hr_reg_write(ud_sq_wqe, UD_SEND_WQE_SL, ah->av.sl);
451 
452 	ud_sq_wqe->sgid_index = ah->av.gid_index;
453 
454 	memcpy(ud_sq_wqe->dmac, ah->av.mac, ETH_ALEN);
455 	memcpy(ud_sq_wqe->dgid, ah->av.dgid, GID_LEN_V2);
456 
457 	if (hr_dev->pci_dev->revision >= PCI_REVISION_ID_HIP09)
458 		return 0;
459 
460 	hr_reg_write(ud_sq_wqe, UD_SEND_WQE_VLAN_EN, ah->av.vlan_en);
461 	hr_reg_write(ud_sq_wqe, UD_SEND_WQE_VLAN, ah->av.vlan_id);
462 
463 	return 0;
464 }
465 
466 static inline int set_ud_wqe(struct hns_roce_qp *qp,
467 			     const struct ib_send_wr *wr,
468 			     void *wqe, unsigned int *sge_idx,
469 			     unsigned int owner_bit)
470 {
471 	struct hns_roce_ah *ah = to_hr_ah(ud_wr(wr)->ah);
472 	struct hns_roce_v2_ud_send_wqe *ud_sq_wqe = wqe;
473 	unsigned int curr_idx = *sge_idx;
474 	unsigned int valid_num_sge;
475 	u32 msg_len = 0;
476 	int ret;
477 
478 	valid_num_sge = calc_wr_sge_num(wr, &msg_len);
479 
480 	ret = set_ud_opcode(ud_sq_wqe, wr);
481 	if (WARN_ON(ret))
482 		return ret;
483 
484 	ud_sq_wqe->msg_len = cpu_to_le32(msg_len);
485 
486 	hr_reg_write(ud_sq_wqe, UD_SEND_WQE_CQE,
487 		     !!(wr->send_flags & IB_SEND_SIGNALED));
488 	hr_reg_write(ud_sq_wqe, UD_SEND_WQE_SE,
489 		     !!(wr->send_flags & IB_SEND_SOLICITED));
490 
491 	hr_reg_write(ud_sq_wqe, UD_SEND_WQE_PD, to_hr_pd(qp->ibqp.pd)->pdn);
492 	hr_reg_write(ud_sq_wqe, UD_SEND_WQE_SGE_NUM, valid_num_sge);
493 	hr_reg_write(ud_sq_wqe, UD_SEND_WQE_MSG_START_SGE_IDX,
494 		     curr_idx & (qp->sge.sge_cnt - 1));
495 
496 	ud_sq_wqe->qkey = cpu_to_le32(ud_wr(wr)->remote_qkey & 0x80000000 ?
497 			  qp->qkey : ud_wr(wr)->remote_qkey);
498 	hr_reg_write(ud_sq_wqe, UD_SEND_WQE_DQPN, ud_wr(wr)->remote_qpn);
499 
500 	ret = fill_ud_av(ud_sq_wqe, ah);
501 	if (ret)
502 		return ret;
503 
504 	qp->sl = to_hr_ah(ud_wr(wr)->ah)->av.sl;
505 
506 	set_extend_sge(qp, wr->sg_list, &curr_idx, valid_num_sge);
507 
508 	/*
509 	 * The pipeline can sequentially post all valid WQEs into WQ buffer,
510 	 * including new WQEs waiting for the doorbell to update the PI again.
511 	 * Therefore, the owner bit of WQE MUST be updated after all fields
512 	 * and extSGEs have been written into DDR instead of cache.
513 	 */
514 	if (qp->en_flags & HNS_ROCE_QP_CAP_OWNER_DB)
515 		dma_wmb();
516 
517 	*sge_idx = curr_idx;
518 	hr_reg_write(ud_sq_wqe, UD_SEND_WQE_OWNER, owner_bit);
519 
520 	return 0;
521 }
522 
523 static int set_rc_opcode(struct hns_roce_dev *hr_dev,
524 			 struct hns_roce_v2_rc_send_wqe *rc_sq_wqe,
525 			 const struct ib_send_wr *wr)
526 {
527 	u32 ib_op = wr->opcode;
528 	int ret = 0;
529 
530 	rc_sq_wqe->immtdata = get_immtdata(wr);
531 
532 	switch (ib_op) {
533 	case IB_WR_RDMA_READ:
534 	case IB_WR_RDMA_WRITE:
535 	case IB_WR_RDMA_WRITE_WITH_IMM:
536 		rc_sq_wqe->rkey = cpu_to_le32(rdma_wr(wr)->rkey);
537 		rc_sq_wqe->va = cpu_to_le64(rdma_wr(wr)->remote_addr);
538 		break;
539 	case IB_WR_SEND:
540 	case IB_WR_SEND_WITH_IMM:
541 		break;
542 	case IB_WR_ATOMIC_CMP_AND_SWP:
543 	case IB_WR_ATOMIC_FETCH_AND_ADD:
544 		rc_sq_wqe->rkey = cpu_to_le32(atomic_wr(wr)->rkey);
545 		rc_sq_wqe->va = cpu_to_le64(atomic_wr(wr)->remote_addr);
546 		break;
547 	case IB_WR_REG_MR:
548 		if (hr_dev->pci_dev->revision >= PCI_REVISION_ID_HIP09)
549 			set_frmr_seg(rc_sq_wqe, reg_wr(wr));
550 		else
551 			ret = -EOPNOTSUPP;
552 		break;
553 	case IB_WR_SEND_WITH_INV:
554 		rc_sq_wqe->inv_key = cpu_to_le32(wr->ex.invalidate_rkey);
555 		break;
556 	default:
557 		ret = -EINVAL;
558 	}
559 
560 	if (unlikely(ret))
561 		return ret;
562 
563 	hr_reg_write(rc_sq_wqe, RC_SEND_WQE_OPCODE, to_hr_opcode(ib_op));
564 
565 	return ret;
566 }
567 
568 static inline int set_rc_wqe(struct hns_roce_qp *qp,
569 			     const struct ib_send_wr *wr,
570 			     void *wqe, unsigned int *sge_idx,
571 			     unsigned int owner_bit)
572 {
573 	struct hns_roce_dev *hr_dev = to_hr_dev(qp->ibqp.device);
574 	struct hns_roce_v2_rc_send_wqe *rc_sq_wqe = wqe;
575 	unsigned int curr_idx = *sge_idx;
576 	unsigned int valid_num_sge;
577 	u32 msg_len = 0;
578 	int ret;
579 
580 	valid_num_sge = calc_wr_sge_num(wr, &msg_len);
581 
582 	rc_sq_wqe->msg_len = cpu_to_le32(msg_len);
583 
584 	ret = set_rc_opcode(hr_dev, rc_sq_wqe, wr);
585 	if (WARN_ON(ret))
586 		return ret;
587 
588 	hr_reg_write(rc_sq_wqe, RC_SEND_WQE_FENCE,
589 		     (wr->send_flags & IB_SEND_FENCE) ? 1 : 0);
590 
591 	hr_reg_write(rc_sq_wqe, RC_SEND_WQE_SE,
592 		     (wr->send_flags & IB_SEND_SOLICITED) ? 1 : 0);
593 
594 	hr_reg_write(rc_sq_wqe, RC_SEND_WQE_CQE,
595 		     (wr->send_flags & IB_SEND_SIGNALED) ? 1 : 0);
596 
597 	if (wr->opcode == IB_WR_ATOMIC_CMP_AND_SWP ||
598 	    wr->opcode == IB_WR_ATOMIC_FETCH_AND_ADD) {
599 		if (msg_len != ATOMIC_WR_LEN)
600 			return -EINVAL;
601 		set_atomic_seg(wr, rc_sq_wqe, valid_num_sge);
602 	} else if (wr->opcode != IB_WR_REG_MR) {
603 		ret = set_rwqe_data_seg(&qp->ibqp, wr, rc_sq_wqe,
604 					&curr_idx, valid_num_sge);
605 		if (ret)
606 			return ret;
607 	}
608 
609 	/*
610 	 * The pipeline can sequentially post all valid WQEs into WQ buffer,
611 	 * including new WQEs waiting for the doorbell to update the PI again.
612 	 * Therefore, the owner bit of WQE MUST be updated after all fields
613 	 * and extSGEs have been written into DDR instead of cache.
614 	 */
615 	if (qp->en_flags & HNS_ROCE_QP_CAP_OWNER_DB)
616 		dma_wmb();
617 
618 	*sge_idx = curr_idx;
619 	hr_reg_write(rc_sq_wqe, RC_SEND_WQE_OWNER, owner_bit);
620 
621 	return ret;
622 }
623 
624 static inline void update_sq_db(struct hns_roce_dev *hr_dev,
625 				struct hns_roce_qp *qp)
626 {
627 	if (unlikely(qp->state == IB_QPS_ERR)) {
628 		flush_cqe(hr_dev, qp);
629 	} else {
630 		struct hns_roce_v2_db sq_db = {};
631 
632 		hr_reg_write(&sq_db, DB_TAG, qp->qpn);
633 		hr_reg_write(&sq_db, DB_CMD, HNS_ROCE_V2_SQ_DB);
634 		hr_reg_write(&sq_db, DB_PI, qp->sq.head);
635 		hr_reg_write(&sq_db, DB_SL, qp->sl);
636 
637 		hns_roce_write64(hr_dev, (__le32 *)&sq_db, qp->sq.db_reg);
638 	}
639 }
640 
641 static inline void update_rq_db(struct hns_roce_dev *hr_dev,
642 				struct hns_roce_qp *qp)
643 {
644 	if (unlikely(qp->state == IB_QPS_ERR)) {
645 		flush_cqe(hr_dev, qp);
646 	} else {
647 		if (likely(qp->en_flags & HNS_ROCE_QP_CAP_RQ_RECORD_DB)) {
648 			*qp->rdb.db_record =
649 					qp->rq.head & V2_DB_PRODUCER_IDX_M;
650 		} else {
651 			struct hns_roce_v2_db rq_db = {};
652 
653 			hr_reg_write(&rq_db, DB_TAG, qp->qpn);
654 			hr_reg_write(&rq_db, DB_CMD, HNS_ROCE_V2_RQ_DB);
655 			hr_reg_write(&rq_db, DB_PI, qp->rq.head);
656 
657 			hns_roce_write64(hr_dev, (__le32 *)&rq_db,
658 					 qp->rq.db_reg);
659 		}
660 	}
661 }
662 
663 static void hns_roce_write512(struct hns_roce_dev *hr_dev, u64 *val,
664 			      u64 __iomem *dest)
665 {
666 #define HNS_ROCE_WRITE_TIMES 8
667 	struct hns_roce_v2_priv *priv = (struct hns_roce_v2_priv *)hr_dev->priv;
668 	struct hnae3_handle *handle = priv->handle;
669 	const struct hnae3_ae_ops *ops = handle->ae_algo->ops;
670 	int i;
671 
672 	if (!hr_dev->dis_db && !ops->get_hw_reset_stat(handle))
673 		for (i = 0; i < HNS_ROCE_WRITE_TIMES; i++)
674 			writeq_relaxed(*(val + i), dest + i);
675 }
676 
677 static void write_dwqe(struct hns_roce_dev *hr_dev, struct hns_roce_qp *qp,
678 		       void *wqe)
679 {
680 #define HNS_ROCE_SL_SHIFT 2
681 	struct hns_roce_v2_rc_send_wqe *rc_sq_wqe = wqe;
682 
683 	/* All kinds of DirectWQE have the same header field layout */
684 	hr_reg_enable(rc_sq_wqe, RC_SEND_WQE_FLAG);
685 	hr_reg_write(rc_sq_wqe, RC_SEND_WQE_DB_SL_L, qp->sl);
686 	hr_reg_write(rc_sq_wqe, RC_SEND_WQE_DB_SL_H,
687 		     qp->sl >> HNS_ROCE_SL_SHIFT);
688 	hr_reg_write(rc_sq_wqe, RC_SEND_WQE_WQE_INDEX, qp->sq.head);
689 
690 	hns_roce_write512(hr_dev, wqe, qp->sq.db_reg);
691 }
692 
693 static int hns_roce_v2_post_send(struct ib_qp *ibqp,
694 				 const struct ib_send_wr *wr,
695 				 const struct ib_send_wr **bad_wr)
696 {
697 	struct hns_roce_dev *hr_dev = to_hr_dev(ibqp->device);
698 	struct ib_device *ibdev = &hr_dev->ib_dev;
699 	struct hns_roce_qp *qp = to_hr_qp(ibqp);
700 	unsigned long flags = 0;
701 	unsigned int owner_bit;
702 	unsigned int sge_idx;
703 	unsigned int wqe_idx;
704 	void *wqe = NULL;
705 	u32 nreq;
706 	int ret;
707 
708 	spin_lock_irqsave(&qp->sq.lock, flags);
709 
710 	ret = check_send_valid(hr_dev, qp);
711 	if (unlikely(ret)) {
712 		*bad_wr = wr;
713 		nreq = 0;
714 		goto out;
715 	}
716 
717 	sge_idx = qp->next_sge;
718 
719 	for (nreq = 0; wr; ++nreq, wr = wr->next) {
720 		if (hns_roce_wq_overflow(&qp->sq, nreq, qp->ibqp.send_cq)) {
721 			ret = -ENOMEM;
722 			*bad_wr = wr;
723 			goto out;
724 		}
725 
726 		wqe_idx = (qp->sq.head + nreq) & (qp->sq.wqe_cnt - 1);
727 
728 		if (unlikely(wr->num_sge > qp->sq.max_gs)) {
729 			ibdev_err(ibdev, "num_sge = %d > qp->sq.max_gs = %u.\n",
730 				  wr->num_sge, qp->sq.max_gs);
731 			ret = -EINVAL;
732 			*bad_wr = wr;
733 			goto out;
734 		}
735 
736 		wqe = hns_roce_get_send_wqe(qp, wqe_idx);
737 		qp->sq.wrid[wqe_idx] = wr->wr_id;
738 		owner_bit =
739 		       ~(((qp->sq.head + nreq) >> ilog2(qp->sq.wqe_cnt)) & 0x1);
740 
741 		/* Corresponding to the QP type, wqe process separately */
742 		if (ibqp->qp_type == IB_QPT_RC)
743 			ret = set_rc_wqe(qp, wr, wqe, &sge_idx, owner_bit);
744 		else
745 			ret = set_ud_wqe(qp, wr, wqe, &sge_idx, owner_bit);
746 
747 		if (unlikely(ret)) {
748 			*bad_wr = wr;
749 			goto out;
750 		}
751 	}
752 
753 out:
754 	if (likely(nreq)) {
755 		qp->sq.head += nreq;
756 		qp->next_sge = sge_idx;
757 
758 		if (nreq == 1 && !ret &&
759 		    (qp->en_flags & HNS_ROCE_QP_CAP_DIRECT_WQE))
760 			write_dwqe(hr_dev, qp, wqe);
761 		else
762 			update_sq_db(hr_dev, qp);
763 	}
764 
765 	spin_unlock_irqrestore(&qp->sq.lock, flags);
766 
767 	return ret;
768 }
769 
770 static int check_recv_valid(struct hns_roce_dev *hr_dev,
771 			    struct hns_roce_qp *hr_qp)
772 {
773 	if (unlikely(hr_dev->state >= HNS_ROCE_DEVICE_STATE_RST_DOWN))
774 		return -EIO;
775 
776 	if (hr_qp->state == IB_QPS_RESET)
777 		return -EINVAL;
778 
779 	return 0;
780 }
781 
782 static void fill_recv_sge_to_wqe(const struct ib_recv_wr *wr, void *wqe,
783 				 u32 max_sge, bool rsv)
784 {
785 	struct hns_roce_v2_wqe_data_seg *dseg = wqe;
786 	u32 i, cnt;
787 
788 	for (i = 0, cnt = 0; i < wr->num_sge; i++) {
789 		/* Skip zero-length sge */
790 		if (!wr->sg_list[i].length)
791 			continue;
792 		set_data_seg_v2(dseg + cnt, wr->sg_list + i);
793 		cnt++;
794 	}
795 
796 	/* Fill a reserved sge to make hw stop reading remaining segments */
797 	if (rsv) {
798 		dseg[cnt].lkey = cpu_to_le32(HNS_ROCE_INVALID_LKEY);
799 		dseg[cnt].addr = 0;
800 		dseg[cnt].len = cpu_to_le32(HNS_ROCE_INVALID_SGE_LENGTH);
801 	} else {
802 		/* Clear remaining segments to make ROCEE ignore sges */
803 		if (cnt < max_sge)
804 			memset(dseg + cnt, 0,
805 			       (max_sge - cnt) * HNS_ROCE_SGE_SIZE);
806 	}
807 }
808 
809 static void fill_rq_wqe(struct hns_roce_qp *hr_qp, const struct ib_recv_wr *wr,
810 			u32 wqe_idx, u32 max_sge)
811 {
812 	void *wqe = NULL;
813 
814 	wqe = hns_roce_get_recv_wqe(hr_qp, wqe_idx);
815 	fill_recv_sge_to_wqe(wr, wqe, max_sge, hr_qp->rq.rsv_sge);
816 }
817 
818 static int hns_roce_v2_post_recv(struct ib_qp *ibqp,
819 				 const struct ib_recv_wr *wr,
820 				 const struct ib_recv_wr **bad_wr)
821 {
822 	struct hns_roce_dev *hr_dev = to_hr_dev(ibqp->device);
823 	struct hns_roce_qp *hr_qp = to_hr_qp(ibqp);
824 	struct ib_device *ibdev = &hr_dev->ib_dev;
825 	u32 wqe_idx, nreq, max_sge;
826 	unsigned long flags;
827 	int ret;
828 
829 	spin_lock_irqsave(&hr_qp->rq.lock, flags);
830 
831 	ret = check_recv_valid(hr_dev, hr_qp);
832 	if (unlikely(ret)) {
833 		*bad_wr = wr;
834 		nreq = 0;
835 		goto out;
836 	}
837 
838 	max_sge = hr_qp->rq.max_gs - hr_qp->rq.rsv_sge;
839 	for (nreq = 0; wr; ++nreq, wr = wr->next) {
840 		if (unlikely(hns_roce_wq_overflow(&hr_qp->rq, nreq,
841 						  hr_qp->ibqp.recv_cq))) {
842 			ret = -ENOMEM;
843 			*bad_wr = wr;
844 			goto out;
845 		}
846 
847 		if (unlikely(wr->num_sge > max_sge)) {
848 			ibdev_err(ibdev, "num_sge = %d >= max_sge = %u.\n",
849 				  wr->num_sge, max_sge);
850 			ret = -EINVAL;
851 			*bad_wr = wr;
852 			goto out;
853 		}
854 
855 		wqe_idx = (hr_qp->rq.head + nreq) & (hr_qp->rq.wqe_cnt - 1);
856 		fill_rq_wqe(hr_qp, wr, wqe_idx, max_sge);
857 		hr_qp->rq.wrid[wqe_idx] = wr->wr_id;
858 	}
859 
860 out:
861 	if (likely(nreq)) {
862 		hr_qp->rq.head += nreq;
863 
864 		update_rq_db(hr_dev, hr_qp);
865 	}
866 	spin_unlock_irqrestore(&hr_qp->rq.lock, flags);
867 
868 	return ret;
869 }
870 
871 static void *get_srq_wqe_buf(struct hns_roce_srq *srq, u32 n)
872 {
873 	return hns_roce_buf_offset(srq->buf_mtr.kmem, n << srq->wqe_shift);
874 }
875 
876 static void *get_idx_buf(struct hns_roce_idx_que *idx_que, u32 n)
877 {
878 	return hns_roce_buf_offset(idx_que->mtr.kmem,
879 				   n << idx_que->entry_shift);
880 }
881 
882 static void hns_roce_free_srq_wqe(struct hns_roce_srq *srq, u32 wqe_index)
883 {
884 	/* always called with interrupts disabled. */
885 	spin_lock(&srq->lock);
886 
887 	bitmap_clear(srq->idx_que.bitmap, wqe_index, 1);
888 	srq->idx_que.tail++;
889 
890 	spin_unlock(&srq->lock);
891 }
892 
893 static int hns_roce_srqwq_overflow(struct hns_roce_srq *srq)
894 {
895 	struct hns_roce_idx_que *idx_que = &srq->idx_que;
896 
897 	return idx_que->head - idx_que->tail >= srq->wqe_cnt;
898 }
899 
900 static int check_post_srq_valid(struct hns_roce_srq *srq, u32 max_sge,
901 				const struct ib_recv_wr *wr)
902 {
903 	struct ib_device *ib_dev = srq->ibsrq.device;
904 
905 	if (unlikely(wr->num_sge > max_sge)) {
906 		ibdev_err(ib_dev,
907 			  "failed to check sge, wr->num_sge = %d, max_sge = %u.\n",
908 			  wr->num_sge, max_sge);
909 		return -EINVAL;
910 	}
911 
912 	if (unlikely(hns_roce_srqwq_overflow(srq))) {
913 		ibdev_err(ib_dev,
914 			  "failed to check srqwq status, srqwq is full.\n");
915 		return -ENOMEM;
916 	}
917 
918 	return 0;
919 }
920 
921 static int get_srq_wqe_idx(struct hns_roce_srq *srq, u32 *wqe_idx)
922 {
923 	struct hns_roce_idx_que *idx_que = &srq->idx_que;
924 	u32 pos;
925 
926 	pos = find_first_zero_bit(idx_que->bitmap, srq->wqe_cnt);
927 	if (unlikely(pos == srq->wqe_cnt))
928 		return -ENOSPC;
929 
930 	bitmap_set(idx_que->bitmap, pos, 1);
931 	*wqe_idx = pos;
932 	return 0;
933 }
934 
935 static void fill_wqe_idx(struct hns_roce_srq *srq, unsigned int wqe_idx)
936 {
937 	struct hns_roce_idx_que *idx_que = &srq->idx_que;
938 	unsigned int head;
939 	__le32 *buf;
940 
941 	head = idx_que->head & (srq->wqe_cnt - 1);
942 
943 	buf = get_idx_buf(idx_que, head);
944 	*buf = cpu_to_le32(wqe_idx);
945 
946 	idx_que->head++;
947 }
948 
949 static void update_srq_db(struct hns_roce_v2_db *db, struct hns_roce_srq *srq)
950 {
951 	hr_reg_write(db, DB_TAG, srq->srqn);
952 	hr_reg_write(db, DB_CMD, HNS_ROCE_V2_SRQ_DB);
953 	hr_reg_write(db, DB_PI, srq->idx_que.head);
954 }
955 
956 static int hns_roce_v2_post_srq_recv(struct ib_srq *ibsrq,
957 				     const struct ib_recv_wr *wr,
958 				     const struct ib_recv_wr **bad_wr)
959 {
960 	struct hns_roce_dev *hr_dev = to_hr_dev(ibsrq->device);
961 	struct hns_roce_srq *srq = to_hr_srq(ibsrq);
962 	struct hns_roce_v2_db srq_db;
963 	unsigned long flags;
964 	int ret = 0;
965 	u32 max_sge;
966 	u32 wqe_idx;
967 	void *wqe;
968 	u32 nreq;
969 
970 	spin_lock_irqsave(&srq->lock, flags);
971 
972 	max_sge = srq->max_gs - srq->rsv_sge;
973 	for (nreq = 0; wr; ++nreq, wr = wr->next) {
974 		ret = check_post_srq_valid(srq, max_sge, wr);
975 		if (ret) {
976 			*bad_wr = wr;
977 			break;
978 		}
979 
980 		ret = get_srq_wqe_idx(srq, &wqe_idx);
981 		if (unlikely(ret)) {
982 			*bad_wr = wr;
983 			break;
984 		}
985 
986 		wqe = get_srq_wqe_buf(srq, wqe_idx);
987 		fill_recv_sge_to_wqe(wr, wqe, max_sge, srq->rsv_sge);
988 		fill_wqe_idx(srq, wqe_idx);
989 		srq->wrid[wqe_idx] = wr->wr_id;
990 	}
991 
992 	if (likely(nreq)) {
993 		update_srq_db(&srq_db, srq);
994 
995 		hns_roce_write64(hr_dev, (__le32 *)&srq_db, srq->db_reg);
996 	}
997 
998 	spin_unlock_irqrestore(&srq->lock, flags);
999 
1000 	return ret;
1001 }
1002 
1003 static u32 hns_roce_v2_cmd_hw_reseted(struct hns_roce_dev *hr_dev,
1004 				      unsigned long instance_stage,
1005 				      unsigned long reset_stage)
1006 {
1007 	/* When hardware reset has been completed once or more, we should stop
1008 	 * sending mailbox&cmq&doorbell to hardware. If now in .init_instance()
1009 	 * function, we should exit with error. If now at HNAE3_INIT_CLIENT
1010 	 * stage of soft reset process, we should exit with error, and then
1011 	 * HNAE3_INIT_CLIENT related process can rollback the operation like
1012 	 * notifing hardware to free resources, HNAE3_INIT_CLIENT related
1013 	 * process will exit with error to notify NIC driver to reschedule soft
1014 	 * reset process once again.
1015 	 */
1016 	hr_dev->is_reset = true;
1017 	hr_dev->dis_db = true;
1018 
1019 	if (reset_stage == HNS_ROCE_STATE_RST_INIT ||
1020 	    instance_stage == HNS_ROCE_STATE_INIT)
1021 		return CMD_RST_PRC_EBUSY;
1022 
1023 	return CMD_RST_PRC_SUCCESS;
1024 }
1025 
1026 static u32 hns_roce_v2_cmd_hw_resetting(struct hns_roce_dev *hr_dev,
1027 					unsigned long instance_stage,
1028 					unsigned long reset_stage)
1029 {
1030 #define HW_RESET_TIMEOUT_US 1000000
1031 #define HW_RESET_SLEEP_US 1000
1032 
1033 	struct hns_roce_v2_priv *priv = hr_dev->priv;
1034 	struct hnae3_handle *handle = priv->handle;
1035 	const struct hnae3_ae_ops *ops = handle->ae_algo->ops;
1036 	unsigned long val;
1037 	int ret;
1038 
1039 	/* When hardware reset is detected, we should stop sending mailbox&cmq&
1040 	 * doorbell to hardware. If now in .init_instance() function, we should
1041 	 * exit with error. If now at HNAE3_INIT_CLIENT stage of soft reset
1042 	 * process, we should exit with error, and then HNAE3_INIT_CLIENT
1043 	 * related process can rollback the operation like notifing hardware to
1044 	 * free resources, HNAE3_INIT_CLIENT related process will exit with
1045 	 * error to notify NIC driver to reschedule soft reset process once
1046 	 * again.
1047 	 */
1048 	hr_dev->dis_db = true;
1049 
1050 	ret = read_poll_timeout(ops->ae_dev_reset_cnt, val,
1051 				val > hr_dev->reset_cnt, HW_RESET_SLEEP_US,
1052 				HW_RESET_TIMEOUT_US, false, handle);
1053 	if (!ret)
1054 		hr_dev->is_reset = true;
1055 
1056 	if (!hr_dev->is_reset || reset_stage == HNS_ROCE_STATE_RST_INIT ||
1057 	    instance_stage == HNS_ROCE_STATE_INIT)
1058 		return CMD_RST_PRC_EBUSY;
1059 
1060 	return CMD_RST_PRC_SUCCESS;
1061 }
1062 
1063 static u32 hns_roce_v2_cmd_sw_resetting(struct hns_roce_dev *hr_dev)
1064 {
1065 	struct hns_roce_v2_priv *priv = hr_dev->priv;
1066 	struct hnae3_handle *handle = priv->handle;
1067 	const struct hnae3_ae_ops *ops = handle->ae_algo->ops;
1068 
1069 	/* When software reset is detected at .init_instance() function, we
1070 	 * should stop sending mailbox&cmq&doorbell to hardware, and exit
1071 	 * with error.
1072 	 */
1073 	hr_dev->dis_db = true;
1074 	if (ops->ae_dev_reset_cnt(handle) != hr_dev->reset_cnt)
1075 		hr_dev->is_reset = true;
1076 
1077 	return CMD_RST_PRC_EBUSY;
1078 }
1079 
1080 static u32 check_aedev_reset_status(struct hns_roce_dev *hr_dev,
1081 				    struct hnae3_handle *handle)
1082 {
1083 	const struct hnae3_ae_ops *ops = handle->ae_algo->ops;
1084 	unsigned long instance_stage; /* the current instance stage */
1085 	unsigned long reset_stage; /* the current reset stage */
1086 	unsigned long reset_cnt;
1087 	bool sw_resetting;
1088 	bool hw_resetting;
1089 
1090 	/* Get information about reset from NIC driver or RoCE driver itself,
1091 	 * the meaning of the following variables from NIC driver are described
1092 	 * as below:
1093 	 * reset_cnt -- The count value of completed hardware reset.
1094 	 * hw_resetting -- Whether hardware device is resetting now.
1095 	 * sw_resetting -- Whether NIC's software reset process is running now.
1096 	 */
1097 	instance_stage = handle->rinfo.instance_state;
1098 	reset_stage = handle->rinfo.reset_state;
1099 	reset_cnt = ops->ae_dev_reset_cnt(handle);
1100 	if (reset_cnt != hr_dev->reset_cnt)
1101 		return hns_roce_v2_cmd_hw_reseted(hr_dev, instance_stage,
1102 						  reset_stage);
1103 
1104 	hw_resetting = ops->get_cmdq_stat(handle);
1105 	if (hw_resetting)
1106 		return hns_roce_v2_cmd_hw_resetting(hr_dev, instance_stage,
1107 						    reset_stage);
1108 
1109 	sw_resetting = ops->ae_dev_resetting(handle);
1110 	if (sw_resetting && instance_stage == HNS_ROCE_STATE_INIT)
1111 		return hns_roce_v2_cmd_sw_resetting(hr_dev);
1112 
1113 	return CMD_RST_PRC_OTHERS;
1114 }
1115 
1116 static bool check_device_is_in_reset(struct hns_roce_dev *hr_dev)
1117 {
1118 	struct hns_roce_v2_priv *priv = hr_dev->priv;
1119 	struct hnae3_handle *handle = priv->handle;
1120 	const struct hnae3_ae_ops *ops = handle->ae_algo->ops;
1121 
1122 	if (hr_dev->reset_cnt != ops->ae_dev_reset_cnt(handle))
1123 		return true;
1124 
1125 	if (ops->get_hw_reset_stat(handle))
1126 		return true;
1127 
1128 	if (ops->ae_dev_resetting(handle))
1129 		return true;
1130 
1131 	return false;
1132 }
1133 
1134 static bool v2_chk_mbox_is_avail(struct hns_roce_dev *hr_dev, bool *busy)
1135 {
1136 	struct hns_roce_v2_priv *priv = hr_dev->priv;
1137 	u32 status;
1138 
1139 	if (hr_dev->is_reset)
1140 		status = CMD_RST_PRC_SUCCESS;
1141 	else
1142 		status = check_aedev_reset_status(hr_dev, priv->handle);
1143 
1144 	*busy = (status == CMD_RST_PRC_EBUSY);
1145 
1146 	return status == CMD_RST_PRC_OTHERS;
1147 }
1148 
1149 static int hns_roce_alloc_cmq_desc(struct hns_roce_dev *hr_dev,
1150 				   struct hns_roce_v2_cmq_ring *ring)
1151 {
1152 	int size = ring->desc_num * sizeof(struct hns_roce_cmq_desc);
1153 
1154 	ring->desc = dma_alloc_coherent(hr_dev->dev, size,
1155 					&ring->desc_dma_addr, GFP_KERNEL);
1156 	if (!ring->desc)
1157 		return -ENOMEM;
1158 
1159 	return 0;
1160 }
1161 
1162 static void hns_roce_free_cmq_desc(struct hns_roce_dev *hr_dev,
1163 				   struct hns_roce_v2_cmq_ring *ring)
1164 {
1165 	dma_free_coherent(hr_dev->dev,
1166 			  ring->desc_num * sizeof(struct hns_roce_cmq_desc),
1167 			  ring->desc, ring->desc_dma_addr);
1168 
1169 	ring->desc_dma_addr = 0;
1170 }
1171 
1172 static int init_csq(struct hns_roce_dev *hr_dev,
1173 		    struct hns_roce_v2_cmq_ring *csq)
1174 {
1175 	dma_addr_t dma;
1176 	int ret;
1177 
1178 	csq->desc_num = CMD_CSQ_DESC_NUM;
1179 	spin_lock_init(&csq->lock);
1180 	csq->flag = TYPE_CSQ;
1181 	csq->head = 0;
1182 
1183 	ret = hns_roce_alloc_cmq_desc(hr_dev, csq);
1184 	if (ret)
1185 		return ret;
1186 
1187 	dma = csq->desc_dma_addr;
1188 	roce_write(hr_dev, ROCEE_TX_CMQ_BASEADDR_L_REG, lower_32_bits(dma));
1189 	roce_write(hr_dev, ROCEE_TX_CMQ_BASEADDR_H_REG, upper_32_bits(dma));
1190 	roce_write(hr_dev, ROCEE_TX_CMQ_DEPTH_REG,
1191 		   (u32)csq->desc_num >> HNS_ROCE_CMQ_DESC_NUM_S);
1192 
1193 	/* Make sure to write CI first and then PI */
1194 	roce_write(hr_dev, ROCEE_TX_CMQ_CI_REG, 0);
1195 	roce_write(hr_dev, ROCEE_TX_CMQ_PI_REG, 0);
1196 
1197 	return 0;
1198 }
1199 
1200 static int hns_roce_v2_cmq_init(struct hns_roce_dev *hr_dev)
1201 {
1202 	struct hns_roce_v2_priv *priv = hr_dev->priv;
1203 	int ret;
1204 
1205 	priv->cmq.tx_timeout = HNS_ROCE_CMQ_TX_TIMEOUT;
1206 
1207 	ret = init_csq(hr_dev, &priv->cmq.csq);
1208 	if (ret)
1209 		dev_err(hr_dev->dev, "failed to init CSQ, ret = %d.\n", ret);
1210 
1211 	return ret;
1212 }
1213 
1214 static void hns_roce_v2_cmq_exit(struct hns_roce_dev *hr_dev)
1215 {
1216 	struct hns_roce_v2_priv *priv = hr_dev->priv;
1217 
1218 	hns_roce_free_cmq_desc(hr_dev, &priv->cmq.csq);
1219 }
1220 
1221 static void hns_roce_cmq_setup_basic_desc(struct hns_roce_cmq_desc *desc,
1222 					  enum hns_roce_opcode_type opcode,
1223 					  bool is_read)
1224 {
1225 	memset((void *)desc, 0, sizeof(struct hns_roce_cmq_desc));
1226 	desc->opcode = cpu_to_le16(opcode);
1227 	desc->flag = cpu_to_le16(HNS_ROCE_CMD_FLAG_IN);
1228 	if (is_read)
1229 		desc->flag |= cpu_to_le16(HNS_ROCE_CMD_FLAG_WR);
1230 	else
1231 		desc->flag &= cpu_to_le16(~HNS_ROCE_CMD_FLAG_WR);
1232 }
1233 
1234 static int hns_roce_cmq_csq_done(struct hns_roce_dev *hr_dev)
1235 {
1236 	u32 tail = roce_read(hr_dev, ROCEE_TX_CMQ_CI_REG);
1237 	struct hns_roce_v2_priv *priv = hr_dev->priv;
1238 
1239 	return tail == priv->cmq.csq.head;
1240 }
1241 
1242 static void update_cmdq_status(struct hns_roce_dev *hr_dev)
1243 {
1244 	struct hns_roce_v2_priv *priv = hr_dev->priv;
1245 	struct hnae3_handle *handle = priv->handle;
1246 
1247 	if (handle->rinfo.reset_state == HNS_ROCE_STATE_RST_INIT ||
1248 	    handle->rinfo.instance_state == HNS_ROCE_STATE_INIT)
1249 		hr_dev->cmd.state = HNS_ROCE_CMDQ_STATE_FATAL_ERR;
1250 }
1251 
1252 static int hns_roce_cmd_err_convert_errno(u16 desc_ret)
1253 {
1254 	struct hns_roce_cmd_errcode errcode_table[] = {
1255 		{CMD_EXEC_SUCCESS, 0},
1256 		{CMD_NO_AUTH, -EPERM},
1257 		{CMD_NOT_EXIST, -EOPNOTSUPP},
1258 		{CMD_CRQ_FULL, -EXFULL},
1259 		{CMD_NEXT_ERR, -ENOSR},
1260 		{CMD_NOT_EXEC, -ENOTBLK},
1261 		{CMD_PARA_ERR, -EINVAL},
1262 		{CMD_RESULT_ERR, -ERANGE},
1263 		{CMD_TIMEOUT, -ETIME},
1264 		{CMD_HILINK_ERR, -ENOLINK},
1265 		{CMD_INFO_ILLEGAL, -ENXIO},
1266 		{CMD_INVALID, -EBADR},
1267 	};
1268 	u16 i;
1269 
1270 	for (i = 0; i < ARRAY_SIZE(errcode_table); i++)
1271 		if (desc_ret == errcode_table[i].return_status)
1272 			return errcode_table[i].errno;
1273 	return -EIO;
1274 }
1275 
1276 static int __hns_roce_cmq_send(struct hns_roce_dev *hr_dev,
1277 			       struct hns_roce_cmq_desc *desc, int num)
1278 {
1279 	struct hns_roce_v2_priv *priv = hr_dev->priv;
1280 	struct hns_roce_v2_cmq_ring *csq = &priv->cmq.csq;
1281 	u32 timeout = 0;
1282 	u16 desc_ret;
1283 	u32 tail;
1284 	int ret;
1285 	int i;
1286 
1287 	spin_lock_bh(&csq->lock);
1288 
1289 	tail = csq->head;
1290 
1291 	for (i = 0; i < num; i++) {
1292 		csq->desc[csq->head++] = desc[i];
1293 		if (csq->head == csq->desc_num)
1294 			csq->head = 0;
1295 	}
1296 
1297 	/* Write to hardware */
1298 	roce_write(hr_dev, ROCEE_TX_CMQ_PI_REG, csq->head);
1299 
1300 	do {
1301 		if (hns_roce_cmq_csq_done(hr_dev))
1302 			break;
1303 		udelay(1);
1304 	} while (++timeout < priv->cmq.tx_timeout);
1305 
1306 	if (hns_roce_cmq_csq_done(hr_dev)) {
1307 		ret = 0;
1308 		for (i = 0; i < num; i++) {
1309 			/* check the result of hardware write back */
1310 			desc[i] = csq->desc[tail++];
1311 			if (tail == csq->desc_num)
1312 				tail = 0;
1313 
1314 			desc_ret = le16_to_cpu(desc[i].retval);
1315 			if (likely(desc_ret == CMD_EXEC_SUCCESS))
1316 				continue;
1317 
1318 			dev_err_ratelimited(hr_dev->dev,
1319 					    "Cmdq IO error, opcode = 0x%x, return = 0x%x.\n",
1320 					    desc->opcode, desc_ret);
1321 			ret = hns_roce_cmd_err_convert_errno(desc_ret);
1322 		}
1323 	} else {
1324 		/* FW/HW reset or incorrect number of desc */
1325 		tail = roce_read(hr_dev, ROCEE_TX_CMQ_CI_REG);
1326 		dev_warn(hr_dev->dev, "CMDQ move tail from %u to %u.\n",
1327 			 csq->head, tail);
1328 		csq->head = tail;
1329 
1330 		update_cmdq_status(hr_dev);
1331 
1332 		ret = -EAGAIN;
1333 	}
1334 
1335 	spin_unlock_bh(&csq->lock);
1336 
1337 	return ret;
1338 }
1339 
1340 static int hns_roce_cmq_send(struct hns_roce_dev *hr_dev,
1341 			     struct hns_roce_cmq_desc *desc, int num)
1342 {
1343 	bool busy;
1344 	int ret;
1345 
1346 	if (hr_dev->cmd.state == HNS_ROCE_CMDQ_STATE_FATAL_ERR)
1347 		return -EIO;
1348 
1349 	if (!v2_chk_mbox_is_avail(hr_dev, &busy))
1350 		return busy ? -EBUSY : 0;
1351 
1352 	ret = __hns_roce_cmq_send(hr_dev, desc, num);
1353 	if (ret) {
1354 		if (!v2_chk_mbox_is_avail(hr_dev, &busy))
1355 			return busy ? -EBUSY : 0;
1356 	}
1357 
1358 	return ret;
1359 }
1360 
1361 static int config_hem_ba_to_hw(struct hns_roce_dev *hr_dev,
1362 			       dma_addr_t base_addr, u8 cmd, unsigned long tag)
1363 {
1364 	struct hns_roce_cmd_mailbox *mbox;
1365 	int ret;
1366 
1367 	mbox = hns_roce_alloc_cmd_mailbox(hr_dev);
1368 	if (IS_ERR(mbox))
1369 		return PTR_ERR(mbox);
1370 
1371 	ret = hns_roce_cmd_mbox(hr_dev, base_addr, mbox->dma, cmd, tag);
1372 	hns_roce_free_cmd_mailbox(hr_dev, mbox);
1373 	return ret;
1374 }
1375 
1376 static int hns_roce_cmq_query_hw_info(struct hns_roce_dev *hr_dev)
1377 {
1378 	struct hns_roce_query_version *resp;
1379 	struct hns_roce_cmq_desc desc;
1380 	int ret;
1381 
1382 	hns_roce_cmq_setup_basic_desc(&desc, HNS_ROCE_OPC_QUERY_HW_VER, true);
1383 	ret = hns_roce_cmq_send(hr_dev, &desc, 1);
1384 	if (ret)
1385 		return ret;
1386 
1387 	resp = (struct hns_roce_query_version *)desc.data;
1388 	hr_dev->hw_rev = le16_to_cpu(resp->rocee_hw_version);
1389 	hr_dev->vendor_id = hr_dev->pci_dev->vendor;
1390 
1391 	return 0;
1392 }
1393 
1394 static void func_clr_hw_resetting_state(struct hns_roce_dev *hr_dev,
1395 					struct hnae3_handle *handle)
1396 {
1397 	const struct hnae3_ae_ops *ops = handle->ae_algo->ops;
1398 	unsigned long end;
1399 
1400 	hr_dev->dis_db = true;
1401 
1402 	dev_warn(hr_dev->dev,
1403 		 "func clear is pending, device in resetting state.\n");
1404 	end = HNS_ROCE_V2_HW_RST_TIMEOUT;
1405 	while (end) {
1406 		if (!ops->get_hw_reset_stat(handle)) {
1407 			hr_dev->is_reset = true;
1408 			dev_info(hr_dev->dev,
1409 				 "func clear success after reset.\n");
1410 			return;
1411 		}
1412 		msleep(HNS_ROCE_V2_HW_RST_COMPLETION_WAIT);
1413 		end -= HNS_ROCE_V2_HW_RST_COMPLETION_WAIT;
1414 	}
1415 
1416 	dev_warn(hr_dev->dev, "func clear failed.\n");
1417 }
1418 
1419 static void func_clr_sw_resetting_state(struct hns_roce_dev *hr_dev,
1420 					struct hnae3_handle *handle)
1421 {
1422 	const struct hnae3_ae_ops *ops = handle->ae_algo->ops;
1423 	unsigned long end;
1424 
1425 	hr_dev->dis_db = true;
1426 
1427 	dev_warn(hr_dev->dev,
1428 		 "func clear is pending, device in resetting state.\n");
1429 	end = HNS_ROCE_V2_HW_RST_TIMEOUT;
1430 	while (end) {
1431 		if (ops->ae_dev_reset_cnt(handle) !=
1432 		    hr_dev->reset_cnt) {
1433 			hr_dev->is_reset = true;
1434 			dev_info(hr_dev->dev,
1435 				 "func clear success after sw reset\n");
1436 			return;
1437 		}
1438 		msleep(HNS_ROCE_V2_HW_RST_COMPLETION_WAIT);
1439 		end -= HNS_ROCE_V2_HW_RST_COMPLETION_WAIT;
1440 	}
1441 
1442 	dev_warn(hr_dev->dev, "func clear failed because of unfinished sw reset\n");
1443 }
1444 
1445 static void hns_roce_func_clr_rst_proc(struct hns_roce_dev *hr_dev, int retval,
1446 				       int flag)
1447 {
1448 	struct hns_roce_v2_priv *priv = hr_dev->priv;
1449 	struct hnae3_handle *handle = priv->handle;
1450 	const struct hnae3_ae_ops *ops = handle->ae_algo->ops;
1451 
1452 	if (ops->ae_dev_reset_cnt(handle) != hr_dev->reset_cnt) {
1453 		hr_dev->dis_db = true;
1454 		hr_dev->is_reset = true;
1455 		dev_info(hr_dev->dev, "func clear success after reset.\n");
1456 		return;
1457 	}
1458 
1459 	if (ops->get_hw_reset_stat(handle)) {
1460 		func_clr_hw_resetting_state(hr_dev, handle);
1461 		return;
1462 	}
1463 
1464 	if (ops->ae_dev_resetting(handle) &&
1465 	    handle->rinfo.instance_state == HNS_ROCE_STATE_INIT) {
1466 		func_clr_sw_resetting_state(hr_dev, handle);
1467 		return;
1468 	}
1469 
1470 	if (retval && !flag)
1471 		dev_warn(hr_dev->dev,
1472 			 "func clear read failed, ret = %d.\n", retval);
1473 
1474 	dev_warn(hr_dev->dev, "func clear failed.\n");
1475 }
1476 
1477 static void __hns_roce_function_clear(struct hns_roce_dev *hr_dev, int vf_id)
1478 {
1479 	bool fclr_write_fail_flag = false;
1480 	struct hns_roce_func_clear *resp;
1481 	struct hns_roce_cmq_desc desc;
1482 	unsigned long end;
1483 	int ret = 0;
1484 
1485 	if (check_device_is_in_reset(hr_dev))
1486 		goto out;
1487 
1488 	hns_roce_cmq_setup_basic_desc(&desc, HNS_ROCE_OPC_FUNC_CLEAR, false);
1489 	resp = (struct hns_roce_func_clear *)desc.data;
1490 	resp->rst_funcid_en = cpu_to_le32(vf_id);
1491 
1492 	ret = hns_roce_cmq_send(hr_dev, &desc, 1);
1493 	if (ret) {
1494 		fclr_write_fail_flag = true;
1495 		dev_err(hr_dev->dev, "func clear write failed, ret = %d.\n",
1496 			 ret);
1497 		goto out;
1498 	}
1499 
1500 	msleep(HNS_ROCE_V2_READ_FUNC_CLEAR_FLAG_INTERVAL);
1501 	end = HNS_ROCE_V2_FUNC_CLEAR_TIMEOUT_MSECS;
1502 	while (end) {
1503 		if (check_device_is_in_reset(hr_dev))
1504 			goto out;
1505 		msleep(HNS_ROCE_V2_READ_FUNC_CLEAR_FLAG_FAIL_WAIT);
1506 		end -= HNS_ROCE_V2_READ_FUNC_CLEAR_FLAG_FAIL_WAIT;
1507 
1508 		hns_roce_cmq_setup_basic_desc(&desc, HNS_ROCE_OPC_FUNC_CLEAR,
1509 					      true);
1510 
1511 		resp->rst_funcid_en = cpu_to_le32(vf_id);
1512 		ret = hns_roce_cmq_send(hr_dev, &desc, 1);
1513 		if (ret)
1514 			continue;
1515 
1516 		if (hr_reg_read(resp, FUNC_CLEAR_RST_FUN_DONE)) {
1517 			if (vf_id == 0)
1518 				hr_dev->is_reset = true;
1519 			return;
1520 		}
1521 	}
1522 
1523 out:
1524 	hns_roce_func_clr_rst_proc(hr_dev, ret, fclr_write_fail_flag);
1525 }
1526 
1527 static int hns_roce_free_vf_resource(struct hns_roce_dev *hr_dev, int vf_id)
1528 {
1529 	enum hns_roce_opcode_type opcode = HNS_ROCE_OPC_ALLOC_VF_RES;
1530 	struct hns_roce_cmq_desc desc[2];
1531 	struct hns_roce_cmq_req *req_a;
1532 
1533 	req_a = (struct hns_roce_cmq_req *)desc[0].data;
1534 	hns_roce_cmq_setup_basic_desc(&desc[0], opcode, false);
1535 	desc[0].flag |= cpu_to_le16(HNS_ROCE_CMD_FLAG_NEXT);
1536 	hns_roce_cmq_setup_basic_desc(&desc[1], opcode, false);
1537 	hr_reg_write(req_a, FUNC_RES_A_VF_ID, vf_id);
1538 
1539 	return hns_roce_cmq_send(hr_dev, desc, 2);
1540 }
1541 
1542 static void hns_roce_function_clear(struct hns_roce_dev *hr_dev)
1543 {
1544 	int ret;
1545 	int i;
1546 
1547 	if (hr_dev->cmd.state == HNS_ROCE_CMDQ_STATE_FATAL_ERR)
1548 		return;
1549 
1550 	for (i = hr_dev->func_num - 1; i >= 0; i--) {
1551 		__hns_roce_function_clear(hr_dev, i);
1552 
1553 		if (i == 0)
1554 			continue;
1555 
1556 		ret = hns_roce_free_vf_resource(hr_dev, i);
1557 		if (ret)
1558 			ibdev_err(&hr_dev->ib_dev,
1559 				  "failed to free vf resource, vf_id = %d, ret = %d.\n",
1560 				  i, ret);
1561 	}
1562 }
1563 
1564 static int hns_roce_clear_extdb_list_info(struct hns_roce_dev *hr_dev)
1565 {
1566 	struct hns_roce_cmq_desc desc;
1567 	int ret;
1568 
1569 	hns_roce_cmq_setup_basic_desc(&desc, HNS_ROCE_OPC_CLEAR_EXTDB_LIST_INFO,
1570 				      false);
1571 	ret = hns_roce_cmq_send(hr_dev, &desc, 1);
1572 	if (ret)
1573 		ibdev_err(&hr_dev->ib_dev,
1574 			  "failed to clear extended doorbell info, ret = %d.\n",
1575 			  ret);
1576 
1577 	return ret;
1578 }
1579 
1580 static int hns_roce_query_fw_ver(struct hns_roce_dev *hr_dev)
1581 {
1582 	struct hns_roce_query_fw_info *resp;
1583 	struct hns_roce_cmq_desc desc;
1584 	int ret;
1585 
1586 	hns_roce_cmq_setup_basic_desc(&desc, HNS_QUERY_FW_VER, true);
1587 	ret = hns_roce_cmq_send(hr_dev, &desc, 1);
1588 	if (ret)
1589 		return ret;
1590 
1591 	resp = (struct hns_roce_query_fw_info *)desc.data;
1592 	hr_dev->caps.fw_ver = (u64)(le32_to_cpu(resp->fw_ver));
1593 
1594 	return 0;
1595 }
1596 
1597 static int hns_roce_query_func_info(struct hns_roce_dev *hr_dev)
1598 {
1599 	struct hns_roce_cmq_desc desc;
1600 	int ret;
1601 
1602 	if (hr_dev->pci_dev->revision == PCI_REVISION_ID_HIP08) {
1603 		hr_dev->func_num = 1;
1604 		return 0;
1605 	}
1606 
1607 	hns_roce_cmq_setup_basic_desc(&desc, HNS_ROCE_OPC_QUERY_FUNC_INFO,
1608 				      true);
1609 	ret = hns_roce_cmq_send(hr_dev, &desc, 1);
1610 	if (ret) {
1611 		hr_dev->func_num = 1;
1612 		return ret;
1613 	}
1614 
1615 	hr_dev->func_num = le32_to_cpu(desc.func_info.own_func_num);
1616 	hr_dev->cong_algo_tmpl_id = le32_to_cpu(desc.func_info.own_mac_id);
1617 
1618 	return 0;
1619 }
1620 
1621 static int hns_roce_hw_v2_query_counter(struct hns_roce_dev *hr_dev,
1622 					u64 *stats, u32 port, int *num_counters)
1623 {
1624 #define CNT_PER_DESC 3
1625 	struct hns_roce_cmq_desc *desc;
1626 	int bd_idx, cnt_idx;
1627 	__le64 *cnt_data;
1628 	int desc_num;
1629 	int ret;
1630 	int i;
1631 
1632 	if (port > hr_dev->caps.num_ports)
1633 		return -EINVAL;
1634 
1635 	desc_num = DIV_ROUND_UP(HNS_ROCE_HW_CNT_TOTAL, CNT_PER_DESC);
1636 	desc = kcalloc(desc_num, sizeof(*desc), GFP_KERNEL);
1637 	if (!desc)
1638 		return -ENOMEM;
1639 
1640 	for (i = 0; i < desc_num; i++) {
1641 		hns_roce_cmq_setup_basic_desc(&desc[i],
1642 					      HNS_ROCE_OPC_QUERY_COUNTER, true);
1643 		if (i != desc_num - 1)
1644 			desc[i].flag |= cpu_to_le16(HNS_ROCE_CMD_FLAG_NEXT);
1645 	}
1646 
1647 	ret = hns_roce_cmq_send(hr_dev, desc, desc_num);
1648 	if (ret) {
1649 		ibdev_err(&hr_dev->ib_dev,
1650 			  "failed to get counter, ret = %d.\n", ret);
1651 		goto err_out;
1652 	}
1653 
1654 	for (i = 0; i < HNS_ROCE_HW_CNT_TOTAL && i < *num_counters; i++) {
1655 		bd_idx = i / CNT_PER_DESC;
1656 		if (bd_idx != HNS_ROCE_HW_CNT_TOTAL / CNT_PER_DESC &&
1657 		    !(desc[bd_idx].flag & cpu_to_le16(HNS_ROCE_CMD_FLAG_NEXT)))
1658 			break;
1659 
1660 		cnt_data = (__le64 *)&desc[bd_idx].data[0];
1661 		cnt_idx = i % CNT_PER_DESC;
1662 		stats[i] = le64_to_cpu(cnt_data[cnt_idx]);
1663 	}
1664 	*num_counters = i;
1665 
1666 err_out:
1667 	kfree(desc);
1668 	return ret;
1669 }
1670 
1671 static int hns_roce_config_global_param(struct hns_roce_dev *hr_dev)
1672 {
1673 	struct hns_roce_cmq_desc desc;
1674 	struct hns_roce_cmq_req *req = (struct hns_roce_cmq_req *)desc.data;
1675 	u32 clock_cycles_of_1us;
1676 
1677 	hns_roce_cmq_setup_basic_desc(&desc, HNS_ROCE_OPC_CFG_GLOBAL_PARAM,
1678 				      false);
1679 
1680 	if (hr_dev->pci_dev->revision == PCI_REVISION_ID_HIP08)
1681 		clock_cycles_of_1us = HNS_ROCE_1NS_CFG;
1682 	else
1683 		clock_cycles_of_1us = HNS_ROCE_1US_CFG;
1684 
1685 	hr_reg_write(req, CFG_GLOBAL_PARAM_1US_CYCLES, clock_cycles_of_1us);
1686 	hr_reg_write(req, CFG_GLOBAL_PARAM_UDP_PORT, ROCE_V2_UDP_DPORT);
1687 
1688 	return hns_roce_cmq_send(hr_dev, &desc, 1);
1689 }
1690 
1691 static int load_func_res_caps(struct hns_roce_dev *hr_dev, bool is_vf)
1692 {
1693 	struct hns_roce_cmq_desc desc[2];
1694 	struct hns_roce_cmq_req *r_a = (struct hns_roce_cmq_req *)desc[0].data;
1695 	struct hns_roce_cmq_req *r_b = (struct hns_roce_cmq_req *)desc[1].data;
1696 	struct hns_roce_caps *caps = &hr_dev->caps;
1697 	enum hns_roce_opcode_type opcode;
1698 	u32 func_num;
1699 	int ret;
1700 
1701 	if (is_vf) {
1702 		opcode = HNS_ROCE_OPC_QUERY_VF_RES;
1703 		func_num = 1;
1704 	} else {
1705 		opcode = HNS_ROCE_OPC_QUERY_PF_RES;
1706 		func_num = hr_dev->func_num;
1707 	}
1708 
1709 	hns_roce_cmq_setup_basic_desc(&desc[0], opcode, true);
1710 	desc[0].flag |= cpu_to_le16(HNS_ROCE_CMD_FLAG_NEXT);
1711 	hns_roce_cmq_setup_basic_desc(&desc[1], opcode, true);
1712 
1713 	ret = hns_roce_cmq_send(hr_dev, desc, 2);
1714 	if (ret)
1715 		return ret;
1716 
1717 	caps->qpc_bt_num = hr_reg_read(r_a, FUNC_RES_A_QPC_BT_NUM) / func_num;
1718 	caps->srqc_bt_num = hr_reg_read(r_a, FUNC_RES_A_SRQC_BT_NUM) / func_num;
1719 	caps->cqc_bt_num = hr_reg_read(r_a, FUNC_RES_A_CQC_BT_NUM) / func_num;
1720 	caps->mpt_bt_num = hr_reg_read(r_a, FUNC_RES_A_MPT_BT_NUM) / func_num;
1721 	caps->eqc_bt_num = hr_reg_read(r_a, FUNC_RES_A_EQC_BT_NUM) / func_num;
1722 	caps->smac_bt_num = hr_reg_read(r_b, FUNC_RES_B_SMAC_NUM) / func_num;
1723 	caps->sgid_bt_num = hr_reg_read(r_b, FUNC_RES_B_SGID_NUM) / func_num;
1724 	caps->sccc_bt_num = hr_reg_read(r_b, FUNC_RES_B_SCCC_BT_NUM) / func_num;
1725 
1726 	if (is_vf) {
1727 		caps->sl_num = hr_reg_read(r_b, FUNC_RES_V_QID_NUM) / func_num;
1728 		caps->gmv_bt_num = hr_reg_read(r_b, FUNC_RES_V_GMV_BT_NUM) /
1729 					       func_num;
1730 	} else {
1731 		caps->sl_num = hr_reg_read(r_b, FUNC_RES_B_QID_NUM) / func_num;
1732 		caps->gmv_bt_num = hr_reg_read(r_b, FUNC_RES_B_GMV_BT_NUM) /
1733 					       func_num;
1734 	}
1735 
1736 	return 0;
1737 }
1738 
1739 static int load_pf_timer_res_caps(struct hns_roce_dev *hr_dev)
1740 {
1741 	struct hns_roce_cmq_desc desc;
1742 	struct hns_roce_cmq_req *req = (struct hns_roce_cmq_req *)desc.data;
1743 	struct hns_roce_caps *caps = &hr_dev->caps;
1744 	int ret;
1745 
1746 	hns_roce_cmq_setup_basic_desc(&desc, HNS_ROCE_OPC_QUERY_PF_TIMER_RES,
1747 				      true);
1748 
1749 	ret = hns_roce_cmq_send(hr_dev, &desc, 1);
1750 	if (ret)
1751 		return ret;
1752 
1753 	caps->qpc_timer_bt_num = hr_reg_read(req, PF_TIMER_RES_QPC_ITEM_NUM);
1754 	caps->cqc_timer_bt_num = hr_reg_read(req, PF_TIMER_RES_CQC_ITEM_NUM);
1755 
1756 	return 0;
1757 }
1758 
1759 static int hns_roce_query_pf_resource(struct hns_roce_dev *hr_dev)
1760 {
1761 	struct device *dev = hr_dev->dev;
1762 	int ret;
1763 
1764 	ret = load_func_res_caps(hr_dev, false);
1765 	if (ret) {
1766 		dev_err(dev, "failed to load pf res caps, ret = %d.\n", ret);
1767 		return ret;
1768 	}
1769 
1770 	ret = load_pf_timer_res_caps(hr_dev);
1771 	if (ret)
1772 		dev_err(dev, "failed to load pf timer resource, ret = %d.\n",
1773 			ret);
1774 
1775 	return ret;
1776 }
1777 
1778 static int hns_roce_query_vf_resource(struct hns_roce_dev *hr_dev)
1779 {
1780 	struct device *dev = hr_dev->dev;
1781 	int ret;
1782 
1783 	ret = load_func_res_caps(hr_dev, true);
1784 	if (ret)
1785 		dev_err(dev, "failed to load vf res caps, ret = %d.\n", ret);
1786 
1787 	return ret;
1788 }
1789 
1790 static int __hns_roce_set_vf_switch_param(struct hns_roce_dev *hr_dev,
1791 					  u32 vf_id)
1792 {
1793 	struct hns_roce_vf_switch *swt;
1794 	struct hns_roce_cmq_desc desc;
1795 	int ret;
1796 
1797 	swt = (struct hns_roce_vf_switch *)desc.data;
1798 	hns_roce_cmq_setup_basic_desc(&desc, HNS_SWITCH_PARAMETER_CFG, true);
1799 	swt->rocee_sel |= cpu_to_le32(HNS_ICL_SWITCH_CMD_ROCEE_SEL);
1800 	hr_reg_write(swt, VF_SWITCH_VF_ID, vf_id);
1801 	ret = hns_roce_cmq_send(hr_dev, &desc, 1);
1802 	if (ret)
1803 		return ret;
1804 
1805 	desc.flag = cpu_to_le16(HNS_ROCE_CMD_FLAG_IN);
1806 	desc.flag &= cpu_to_le16(~HNS_ROCE_CMD_FLAG_WR);
1807 	hr_reg_enable(swt, VF_SWITCH_ALW_LPBK);
1808 	hr_reg_clear(swt, VF_SWITCH_ALW_LCL_LPBK);
1809 	hr_reg_enable(swt, VF_SWITCH_ALW_DST_OVRD);
1810 
1811 	return hns_roce_cmq_send(hr_dev, &desc, 1);
1812 }
1813 
1814 static int hns_roce_set_vf_switch_param(struct hns_roce_dev *hr_dev)
1815 {
1816 	u32 vf_id;
1817 	int ret;
1818 
1819 	for (vf_id = 0; vf_id < hr_dev->func_num; vf_id++) {
1820 		ret = __hns_roce_set_vf_switch_param(hr_dev, vf_id);
1821 		if (ret)
1822 			return ret;
1823 	}
1824 	return 0;
1825 }
1826 
1827 static int config_vf_hem_resource(struct hns_roce_dev *hr_dev, int vf_id)
1828 {
1829 	struct hns_roce_cmq_desc desc[2];
1830 	struct hns_roce_cmq_req *r_a = (struct hns_roce_cmq_req *)desc[0].data;
1831 	struct hns_roce_cmq_req *r_b = (struct hns_roce_cmq_req *)desc[1].data;
1832 	enum hns_roce_opcode_type opcode = HNS_ROCE_OPC_ALLOC_VF_RES;
1833 	struct hns_roce_caps *caps = &hr_dev->caps;
1834 
1835 	hns_roce_cmq_setup_basic_desc(&desc[0], opcode, false);
1836 	desc[0].flag |= cpu_to_le16(HNS_ROCE_CMD_FLAG_NEXT);
1837 	hns_roce_cmq_setup_basic_desc(&desc[1], opcode, false);
1838 
1839 	hr_reg_write(r_a, FUNC_RES_A_VF_ID, vf_id);
1840 
1841 	hr_reg_write(r_a, FUNC_RES_A_QPC_BT_NUM, caps->qpc_bt_num);
1842 	hr_reg_write(r_a, FUNC_RES_A_QPC_BT_IDX, vf_id * caps->qpc_bt_num);
1843 	hr_reg_write(r_a, FUNC_RES_A_SRQC_BT_NUM, caps->srqc_bt_num);
1844 	hr_reg_write(r_a, FUNC_RES_A_SRQC_BT_IDX, vf_id * caps->srqc_bt_num);
1845 	hr_reg_write(r_a, FUNC_RES_A_CQC_BT_NUM, caps->cqc_bt_num);
1846 	hr_reg_write(r_a, FUNC_RES_A_CQC_BT_IDX, vf_id * caps->cqc_bt_num);
1847 	hr_reg_write(r_a, FUNC_RES_A_MPT_BT_NUM, caps->mpt_bt_num);
1848 	hr_reg_write(r_a, FUNC_RES_A_MPT_BT_IDX, vf_id * caps->mpt_bt_num);
1849 	hr_reg_write(r_a, FUNC_RES_A_EQC_BT_NUM, caps->eqc_bt_num);
1850 	hr_reg_write(r_a, FUNC_RES_A_EQC_BT_IDX, vf_id * caps->eqc_bt_num);
1851 	hr_reg_write(r_b, FUNC_RES_V_QID_NUM, caps->sl_num);
1852 	hr_reg_write(r_b, FUNC_RES_B_QID_IDX, vf_id * caps->sl_num);
1853 	hr_reg_write(r_b, FUNC_RES_B_SCCC_BT_NUM, caps->sccc_bt_num);
1854 	hr_reg_write(r_b, FUNC_RES_B_SCCC_BT_IDX, vf_id * caps->sccc_bt_num);
1855 
1856 	if (hr_dev->pci_dev->revision >= PCI_REVISION_ID_HIP09) {
1857 		hr_reg_write(r_b, FUNC_RES_V_GMV_BT_NUM, caps->gmv_bt_num);
1858 		hr_reg_write(r_b, FUNC_RES_B_GMV_BT_IDX,
1859 			     vf_id * caps->gmv_bt_num);
1860 	} else {
1861 		hr_reg_write(r_b, FUNC_RES_B_SGID_NUM, caps->sgid_bt_num);
1862 		hr_reg_write(r_b, FUNC_RES_B_SGID_IDX,
1863 			     vf_id * caps->sgid_bt_num);
1864 		hr_reg_write(r_b, FUNC_RES_B_SMAC_NUM, caps->smac_bt_num);
1865 		hr_reg_write(r_b, FUNC_RES_B_SMAC_IDX,
1866 			     vf_id * caps->smac_bt_num);
1867 	}
1868 
1869 	return hns_roce_cmq_send(hr_dev, desc, 2);
1870 }
1871 
1872 static int hns_roce_alloc_vf_resource(struct hns_roce_dev *hr_dev)
1873 {
1874 	u32 func_num = max_t(u32, 1, hr_dev->func_num);
1875 	u32 vf_id;
1876 	int ret;
1877 
1878 	for (vf_id = 0; vf_id < func_num; vf_id++) {
1879 		ret = config_vf_hem_resource(hr_dev, vf_id);
1880 		if (ret) {
1881 			dev_err(hr_dev->dev,
1882 				"failed to config vf-%u hem res, ret = %d.\n",
1883 				vf_id, ret);
1884 			return ret;
1885 		}
1886 	}
1887 
1888 	return 0;
1889 }
1890 
1891 static int hns_roce_v2_set_bt(struct hns_roce_dev *hr_dev)
1892 {
1893 	struct hns_roce_cmq_desc desc;
1894 	struct hns_roce_cmq_req *req = (struct hns_roce_cmq_req *)desc.data;
1895 	struct hns_roce_caps *caps = &hr_dev->caps;
1896 
1897 	hns_roce_cmq_setup_basic_desc(&desc, HNS_ROCE_OPC_CFG_BT_ATTR, false);
1898 
1899 	hr_reg_write(req, CFG_BT_ATTR_QPC_BA_PGSZ,
1900 		     caps->qpc_ba_pg_sz + PG_SHIFT_OFFSET);
1901 	hr_reg_write(req, CFG_BT_ATTR_QPC_BUF_PGSZ,
1902 		     caps->qpc_buf_pg_sz + PG_SHIFT_OFFSET);
1903 	hr_reg_write(req, CFG_BT_ATTR_QPC_HOPNUM,
1904 		     to_hr_hem_hopnum(caps->qpc_hop_num, caps->num_qps));
1905 
1906 	hr_reg_write(req, CFG_BT_ATTR_SRQC_BA_PGSZ,
1907 		     caps->srqc_ba_pg_sz + PG_SHIFT_OFFSET);
1908 	hr_reg_write(req, CFG_BT_ATTR_SRQC_BUF_PGSZ,
1909 		     caps->srqc_buf_pg_sz + PG_SHIFT_OFFSET);
1910 	hr_reg_write(req, CFG_BT_ATTR_SRQC_HOPNUM,
1911 		     to_hr_hem_hopnum(caps->srqc_hop_num, caps->num_srqs));
1912 
1913 	hr_reg_write(req, CFG_BT_ATTR_CQC_BA_PGSZ,
1914 		     caps->cqc_ba_pg_sz + PG_SHIFT_OFFSET);
1915 	hr_reg_write(req, CFG_BT_ATTR_CQC_BUF_PGSZ,
1916 		     caps->cqc_buf_pg_sz + PG_SHIFT_OFFSET);
1917 	hr_reg_write(req, CFG_BT_ATTR_CQC_HOPNUM,
1918 		     to_hr_hem_hopnum(caps->cqc_hop_num, caps->num_cqs));
1919 
1920 	hr_reg_write(req, CFG_BT_ATTR_MPT_BA_PGSZ,
1921 		     caps->mpt_ba_pg_sz + PG_SHIFT_OFFSET);
1922 	hr_reg_write(req, CFG_BT_ATTR_MPT_BUF_PGSZ,
1923 		     caps->mpt_buf_pg_sz + PG_SHIFT_OFFSET);
1924 	hr_reg_write(req, CFG_BT_ATTR_MPT_HOPNUM,
1925 		     to_hr_hem_hopnum(caps->mpt_hop_num, caps->num_mtpts));
1926 
1927 	hr_reg_write(req, CFG_BT_ATTR_SCCC_BA_PGSZ,
1928 		     caps->sccc_ba_pg_sz + PG_SHIFT_OFFSET);
1929 	hr_reg_write(req, CFG_BT_ATTR_SCCC_BUF_PGSZ,
1930 		     caps->sccc_buf_pg_sz + PG_SHIFT_OFFSET);
1931 	hr_reg_write(req, CFG_BT_ATTR_SCCC_HOPNUM,
1932 		     to_hr_hem_hopnum(caps->sccc_hop_num, caps->num_qps));
1933 
1934 	return hns_roce_cmq_send(hr_dev, &desc, 1);
1935 }
1936 
1937 static void calc_pg_sz(u32 obj_num, u32 obj_size, u32 hop_num, u32 ctx_bt_num,
1938 		       u32 *buf_page_size, u32 *bt_page_size, u32 hem_type)
1939 {
1940 	u64 obj_per_chunk;
1941 	u64 bt_chunk_size = PAGE_SIZE;
1942 	u64 buf_chunk_size = PAGE_SIZE;
1943 	u64 obj_per_chunk_default = buf_chunk_size / obj_size;
1944 
1945 	*buf_page_size = 0;
1946 	*bt_page_size = 0;
1947 
1948 	switch (hop_num) {
1949 	case 3:
1950 		obj_per_chunk = ctx_bt_num * (bt_chunk_size / BA_BYTE_LEN) *
1951 				(bt_chunk_size / BA_BYTE_LEN) *
1952 				(bt_chunk_size / BA_BYTE_LEN) *
1953 				 obj_per_chunk_default;
1954 		break;
1955 	case 2:
1956 		obj_per_chunk = ctx_bt_num * (bt_chunk_size / BA_BYTE_LEN) *
1957 				(bt_chunk_size / BA_BYTE_LEN) *
1958 				 obj_per_chunk_default;
1959 		break;
1960 	case 1:
1961 		obj_per_chunk = ctx_bt_num * (bt_chunk_size / BA_BYTE_LEN) *
1962 				obj_per_chunk_default;
1963 		break;
1964 	case HNS_ROCE_HOP_NUM_0:
1965 		obj_per_chunk = ctx_bt_num * obj_per_chunk_default;
1966 		break;
1967 	default:
1968 		pr_err("table %u not support hop_num = %u!\n", hem_type,
1969 		       hop_num);
1970 		return;
1971 	}
1972 
1973 	if (hem_type >= HEM_TYPE_MTT)
1974 		*bt_page_size = ilog2(DIV_ROUND_UP(obj_num, obj_per_chunk));
1975 	else
1976 		*buf_page_size = ilog2(DIV_ROUND_UP(obj_num, obj_per_chunk));
1977 }
1978 
1979 static void set_hem_page_size(struct hns_roce_dev *hr_dev)
1980 {
1981 	struct hns_roce_caps *caps = &hr_dev->caps;
1982 
1983 	/* EQ */
1984 	caps->eqe_ba_pg_sz = 0;
1985 	caps->eqe_buf_pg_sz = 0;
1986 
1987 	/* Link Table */
1988 	caps->llm_buf_pg_sz = 0;
1989 
1990 	/* MR */
1991 	caps->mpt_ba_pg_sz = 0;
1992 	caps->mpt_buf_pg_sz = 0;
1993 	caps->pbl_ba_pg_sz = HNS_ROCE_BA_PG_SZ_SUPPORTED_16K;
1994 	caps->pbl_buf_pg_sz = 0;
1995 	calc_pg_sz(caps->num_mtpts, caps->mtpt_entry_sz, caps->mpt_hop_num,
1996 		   caps->mpt_bt_num, &caps->mpt_buf_pg_sz, &caps->mpt_ba_pg_sz,
1997 		   HEM_TYPE_MTPT);
1998 
1999 	/* QP */
2000 	caps->qpc_ba_pg_sz = 0;
2001 	caps->qpc_buf_pg_sz = 0;
2002 	caps->qpc_timer_ba_pg_sz = 0;
2003 	caps->qpc_timer_buf_pg_sz = 0;
2004 	caps->sccc_ba_pg_sz = 0;
2005 	caps->sccc_buf_pg_sz = 0;
2006 	caps->mtt_ba_pg_sz = 0;
2007 	caps->mtt_buf_pg_sz = 0;
2008 	calc_pg_sz(caps->num_qps, caps->qpc_sz, caps->qpc_hop_num,
2009 		   caps->qpc_bt_num, &caps->qpc_buf_pg_sz, &caps->qpc_ba_pg_sz,
2010 		   HEM_TYPE_QPC);
2011 
2012 	if (caps->flags & HNS_ROCE_CAP_FLAG_QP_FLOW_CTRL)
2013 		calc_pg_sz(caps->num_qps, caps->sccc_sz, caps->sccc_hop_num,
2014 			   caps->sccc_bt_num, &caps->sccc_buf_pg_sz,
2015 			   &caps->sccc_ba_pg_sz, HEM_TYPE_SCCC);
2016 
2017 	/* CQ */
2018 	caps->cqc_ba_pg_sz = 0;
2019 	caps->cqc_buf_pg_sz = 0;
2020 	caps->cqc_timer_ba_pg_sz = 0;
2021 	caps->cqc_timer_buf_pg_sz = 0;
2022 	caps->cqe_ba_pg_sz = HNS_ROCE_BA_PG_SZ_SUPPORTED_256K;
2023 	caps->cqe_buf_pg_sz = 0;
2024 	calc_pg_sz(caps->num_cqs, caps->cqc_entry_sz, caps->cqc_hop_num,
2025 		   caps->cqc_bt_num, &caps->cqc_buf_pg_sz, &caps->cqc_ba_pg_sz,
2026 		   HEM_TYPE_CQC);
2027 	calc_pg_sz(caps->max_cqes, caps->cqe_sz, caps->cqe_hop_num,
2028 		   1, &caps->cqe_buf_pg_sz, &caps->cqe_ba_pg_sz, HEM_TYPE_CQE);
2029 
2030 	/* SRQ */
2031 	if (caps->flags & HNS_ROCE_CAP_FLAG_SRQ) {
2032 		caps->srqc_ba_pg_sz = 0;
2033 		caps->srqc_buf_pg_sz = 0;
2034 		caps->srqwqe_ba_pg_sz = 0;
2035 		caps->srqwqe_buf_pg_sz = 0;
2036 		caps->idx_ba_pg_sz = 0;
2037 		caps->idx_buf_pg_sz = 0;
2038 		calc_pg_sz(caps->num_srqs, caps->srqc_entry_sz,
2039 			   caps->srqc_hop_num, caps->srqc_bt_num,
2040 			   &caps->srqc_buf_pg_sz, &caps->srqc_ba_pg_sz,
2041 			   HEM_TYPE_SRQC);
2042 		calc_pg_sz(caps->num_srqwqe_segs, caps->mtt_entry_sz,
2043 			   caps->srqwqe_hop_num, 1, &caps->srqwqe_buf_pg_sz,
2044 			   &caps->srqwqe_ba_pg_sz, HEM_TYPE_SRQWQE);
2045 		calc_pg_sz(caps->num_idx_segs, caps->idx_entry_sz,
2046 			   caps->idx_hop_num, 1, &caps->idx_buf_pg_sz,
2047 			   &caps->idx_ba_pg_sz, HEM_TYPE_IDX);
2048 	}
2049 
2050 	/* GMV */
2051 	caps->gmv_ba_pg_sz = 0;
2052 	caps->gmv_buf_pg_sz = 0;
2053 }
2054 
2055 /* Apply all loaded caps before setting to hardware */
2056 static void apply_func_caps(struct hns_roce_dev *hr_dev)
2057 {
2058 	struct hns_roce_caps *caps = &hr_dev->caps;
2059 	struct hns_roce_v2_priv *priv = hr_dev->priv;
2060 
2061 	/* The following configurations don't need to be got from firmware. */
2062 	caps->qpc_timer_entry_sz = HNS_ROCE_V2_QPC_TIMER_ENTRY_SZ;
2063 	caps->cqc_timer_entry_sz = HNS_ROCE_V2_CQC_TIMER_ENTRY_SZ;
2064 	caps->mtt_entry_sz = HNS_ROCE_V2_MTT_ENTRY_SZ;
2065 
2066 	caps->pbl_hop_num = HNS_ROCE_PBL_HOP_NUM;
2067 	caps->qpc_timer_hop_num = HNS_ROCE_HOP_NUM_0;
2068 	caps->cqc_timer_hop_num = HNS_ROCE_HOP_NUM_0;
2069 
2070 	caps->num_srqwqe_segs = HNS_ROCE_V2_MAX_SRQWQE_SEGS;
2071 	caps->num_idx_segs = HNS_ROCE_V2_MAX_IDX_SEGS;
2072 
2073 	if (!caps->num_comp_vectors)
2074 		caps->num_comp_vectors =
2075 			min_t(u32, caps->eqc_bt_num - HNS_ROCE_V2_AEQE_VEC_NUM,
2076 				(u32)priv->handle->rinfo.num_vectors -
2077 		(HNS_ROCE_V2_AEQE_VEC_NUM + HNS_ROCE_V2_ABNORMAL_VEC_NUM));
2078 
2079 	if (hr_dev->pci_dev->revision >= PCI_REVISION_ID_HIP09) {
2080 		caps->eqe_hop_num = HNS_ROCE_V3_EQE_HOP_NUM;
2081 		caps->ceqe_size = HNS_ROCE_V3_EQE_SIZE;
2082 		caps->aeqe_size = HNS_ROCE_V3_EQE_SIZE;
2083 
2084 		/* The following configurations will be overwritten */
2085 		caps->qpc_sz = HNS_ROCE_V3_QPC_SZ;
2086 		caps->cqe_sz = HNS_ROCE_V3_CQE_SIZE;
2087 		caps->sccc_sz = HNS_ROCE_V3_SCCC_SZ;
2088 
2089 		/* The following configurations are not got from firmware */
2090 		caps->gmv_entry_sz = HNS_ROCE_V3_GMV_ENTRY_SZ;
2091 
2092 		caps->gmv_hop_num = HNS_ROCE_HOP_NUM_0;
2093 		caps->gid_table_len[0] = caps->gmv_bt_num *
2094 					(HNS_HW_PAGE_SIZE / caps->gmv_entry_sz);
2095 
2096 		caps->gmv_entry_num = caps->gmv_bt_num * (HNS_HW_PAGE_SIZE /
2097 							  caps->gmv_entry_sz);
2098 	} else {
2099 		u32 func_num = max_t(u32, 1, hr_dev->func_num);
2100 
2101 		caps->eqe_hop_num = HNS_ROCE_V2_EQE_HOP_NUM;
2102 		caps->ceqe_size = HNS_ROCE_CEQE_SIZE;
2103 		caps->aeqe_size = HNS_ROCE_AEQE_SIZE;
2104 		caps->gid_table_len[0] /= func_num;
2105 	}
2106 
2107 	if (hr_dev->is_vf) {
2108 		caps->default_aeq_arm_st = 0x3;
2109 		caps->default_ceq_arm_st = 0x3;
2110 		caps->default_ceq_max_cnt = 0x1;
2111 		caps->default_ceq_period = 0x10;
2112 		caps->default_aeq_max_cnt = 0x1;
2113 		caps->default_aeq_period = 0x10;
2114 	}
2115 
2116 	set_hem_page_size(hr_dev);
2117 }
2118 
2119 static int hns_roce_query_caps(struct hns_roce_dev *hr_dev)
2120 {
2121 	struct hns_roce_cmq_desc desc[HNS_ROCE_QUERY_PF_CAPS_CMD_NUM];
2122 	struct hns_roce_caps *caps = &hr_dev->caps;
2123 	struct hns_roce_query_pf_caps_a *resp_a;
2124 	struct hns_roce_query_pf_caps_b *resp_b;
2125 	struct hns_roce_query_pf_caps_c *resp_c;
2126 	struct hns_roce_query_pf_caps_d *resp_d;
2127 	struct hns_roce_query_pf_caps_e *resp_e;
2128 	enum hns_roce_opcode_type cmd;
2129 	int ctx_hop_num;
2130 	int pbl_hop_num;
2131 	int ret;
2132 	int i;
2133 
2134 	cmd = hr_dev->is_vf ? HNS_ROCE_OPC_QUERY_VF_CAPS_NUM :
2135 	      HNS_ROCE_OPC_QUERY_PF_CAPS_NUM;
2136 
2137 	for (i = 0; i < HNS_ROCE_QUERY_PF_CAPS_CMD_NUM; i++) {
2138 		hns_roce_cmq_setup_basic_desc(&desc[i], cmd, true);
2139 		if (i < (HNS_ROCE_QUERY_PF_CAPS_CMD_NUM - 1))
2140 			desc[i].flag |= cpu_to_le16(HNS_ROCE_CMD_FLAG_NEXT);
2141 		else
2142 			desc[i].flag &= ~cpu_to_le16(HNS_ROCE_CMD_FLAG_NEXT);
2143 	}
2144 
2145 	ret = hns_roce_cmq_send(hr_dev, desc, HNS_ROCE_QUERY_PF_CAPS_CMD_NUM);
2146 	if (ret)
2147 		return ret;
2148 
2149 	resp_a = (struct hns_roce_query_pf_caps_a *)desc[0].data;
2150 	resp_b = (struct hns_roce_query_pf_caps_b *)desc[1].data;
2151 	resp_c = (struct hns_roce_query_pf_caps_c *)desc[2].data;
2152 	resp_d = (struct hns_roce_query_pf_caps_d *)desc[3].data;
2153 	resp_e = (struct hns_roce_query_pf_caps_e *)desc[4].data;
2154 
2155 	caps->local_ca_ack_delay = resp_a->local_ca_ack_delay;
2156 	caps->max_sq_sg = le16_to_cpu(resp_a->max_sq_sg);
2157 	caps->max_sq_inline = le16_to_cpu(resp_a->max_sq_inline);
2158 	caps->max_rq_sg = le16_to_cpu(resp_a->max_rq_sg);
2159 	caps->max_rq_sg = roundup_pow_of_two(caps->max_rq_sg);
2160 	caps->max_srq_sges = le16_to_cpu(resp_a->max_srq_sges);
2161 	caps->max_srq_sges = roundup_pow_of_two(caps->max_srq_sges);
2162 	caps->num_aeq_vectors = resp_a->num_aeq_vectors;
2163 	caps->num_other_vectors = resp_a->num_other_vectors;
2164 	caps->max_sq_desc_sz = resp_a->max_sq_desc_sz;
2165 	caps->max_rq_desc_sz = resp_a->max_rq_desc_sz;
2166 
2167 	caps->mtpt_entry_sz = resp_b->mtpt_entry_sz;
2168 	caps->irrl_entry_sz = resp_b->irrl_entry_sz;
2169 	caps->trrl_entry_sz = resp_b->trrl_entry_sz;
2170 	caps->cqc_entry_sz = resp_b->cqc_entry_sz;
2171 	caps->srqc_entry_sz = resp_b->srqc_entry_sz;
2172 	caps->idx_entry_sz = resp_b->idx_entry_sz;
2173 	caps->sccc_sz = resp_b->sccc_sz;
2174 	caps->max_mtu = resp_b->max_mtu;
2175 	caps->min_cqes = resp_b->min_cqes;
2176 	caps->min_wqes = resp_b->min_wqes;
2177 	caps->page_size_cap = le32_to_cpu(resp_b->page_size_cap);
2178 	caps->pkey_table_len[0] = resp_b->pkey_table_len;
2179 	caps->phy_num_uars = resp_b->phy_num_uars;
2180 	ctx_hop_num = resp_b->ctx_hop_num;
2181 	pbl_hop_num = resp_b->pbl_hop_num;
2182 
2183 	caps->num_pds = 1 << hr_reg_read(resp_c, PF_CAPS_C_NUM_PDS);
2184 
2185 	caps->flags = hr_reg_read(resp_c, PF_CAPS_C_CAP_FLAGS);
2186 	caps->flags |= le16_to_cpu(resp_d->cap_flags_ex) <<
2187 		       HNS_ROCE_CAP_FLAGS_EX_SHIFT;
2188 
2189 	caps->num_cqs = 1 << hr_reg_read(resp_c, PF_CAPS_C_NUM_CQS);
2190 	caps->gid_table_len[0] = hr_reg_read(resp_c, PF_CAPS_C_MAX_GID);
2191 	caps->max_cqes = 1 << hr_reg_read(resp_c, PF_CAPS_C_CQ_DEPTH);
2192 	caps->num_xrcds = 1 << hr_reg_read(resp_c, PF_CAPS_C_NUM_XRCDS);
2193 	caps->num_mtpts = 1 << hr_reg_read(resp_c, PF_CAPS_C_NUM_MRWS);
2194 	caps->num_qps = 1 << hr_reg_read(resp_c, PF_CAPS_C_NUM_QPS);
2195 	caps->max_qp_init_rdma = hr_reg_read(resp_c, PF_CAPS_C_MAX_ORD);
2196 	caps->max_qp_dest_rdma = caps->max_qp_init_rdma;
2197 	caps->max_wqes = 1 << le16_to_cpu(resp_c->sq_depth);
2198 
2199 	caps->num_srqs = 1 << hr_reg_read(resp_d, PF_CAPS_D_NUM_SRQS);
2200 	caps->cong_type = hr_reg_read(resp_d, PF_CAPS_D_CONG_TYPE);
2201 	caps->max_srq_wrs = 1 << le16_to_cpu(resp_d->srq_depth);
2202 	caps->ceqe_depth = 1 << hr_reg_read(resp_d, PF_CAPS_D_CEQ_DEPTH);
2203 	caps->num_comp_vectors = hr_reg_read(resp_d, PF_CAPS_D_NUM_CEQS);
2204 	caps->aeqe_depth = 1 << hr_reg_read(resp_d, PF_CAPS_D_AEQ_DEPTH);
2205 	caps->reserved_pds = hr_reg_read(resp_d, PF_CAPS_D_RSV_PDS);
2206 	caps->num_uars = 1 << hr_reg_read(resp_d, PF_CAPS_D_NUM_UARS);
2207 	caps->reserved_qps = hr_reg_read(resp_d, PF_CAPS_D_RSV_QPS);
2208 	caps->reserved_uars = hr_reg_read(resp_d, PF_CAPS_D_RSV_UARS);
2209 
2210 	caps->reserved_mrws = hr_reg_read(resp_e, PF_CAPS_E_RSV_MRWS);
2211 	caps->chunk_sz = 1 << hr_reg_read(resp_e, PF_CAPS_E_CHUNK_SIZE_SHIFT);
2212 	caps->reserved_cqs = hr_reg_read(resp_e, PF_CAPS_E_RSV_CQS);
2213 	caps->reserved_xrcds = hr_reg_read(resp_e, PF_CAPS_E_RSV_XRCDS);
2214 	caps->reserved_srqs = hr_reg_read(resp_e, PF_CAPS_E_RSV_SRQS);
2215 	caps->reserved_lkey = hr_reg_read(resp_e, PF_CAPS_E_RSV_LKEYS);
2216 
2217 	caps->qpc_hop_num = ctx_hop_num;
2218 	caps->sccc_hop_num = ctx_hop_num;
2219 	caps->srqc_hop_num = ctx_hop_num;
2220 	caps->cqc_hop_num = ctx_hop_num;
2221 	caps->mpt_hop_num = ctx_hop_num;
2222 	caps->mtt_hop_num = pbl_hop_num;
2223 	caps->cqe_hop_num = pbl_hop_num;
2224 	caps->srqwqe_hop_num = pbl_hop_num;
2225 	caps->idx_hop_num = pbl_hop_num;
2226 	caps->wqe_sq_hop_num = hr_reg_read(resp_d, PF_CAPS_D_SQWQE_HOP_NUM);
2227 	caps->wqe_sge_hop_num = hr_reg_read(resp_d, PF_CAPS_D_EX_SGE_HOP_NUM);
2228 	caps->wqe_rq_hop_num = hr_reg_read(resp_d, PF_CAPS_D_RQWQE_HOP_NUM);
2229 
2230 	if (!(caps->page_size_cap & PAGE_SIZE))
2231 		caps->page_size_cap = HNS_ROCE_V2_PAGE_SIZE_SUPPORTED;
2232 
2233 	if (!hr_dev->is_vf) {
2234 		caps->cqe_sz = resp_a->cqe_sz;
2235 		caps->qpc_sz = le16_to_cpu(resp_b->qpc_sz);
2236 		caps->default_aeq_arm_st =
2237 				hr_reg_read(resp_d, PF_CAPS_D_AEQ_ARM_ST);
2238 		caps->default_ceq_arm_st =
2239 				hr_reg_read(resp_d, PF_CAPS_D_CEQ_ARM_ST);
2240 		caps->default_ceq_max_cnt = le16_to_cpu(resp_e->ceq_max_cnt);
2241 		caps->default_ceq_period = le16_to_cpu(resp_e->ceq_period);
2242 		caps->default_aeq_max_cnt = le16_to_cpu(resp_e->aeq_max_cnt);
2243 		caps->default_aeq_period = le16_to_cpu(resp_e->aeq_period);
2244 	}
2245 
2246 	return 0;
2247 }
2248 
2249 static int config_hem_entry_size(struct hns_roce_dev *hr_dev, u32 type, u32 val)
2250 {
2251 	struct hns_roce_cmq_desc desc;
2252 	struct hns_roce_cmq_req *req = (struct hns_roce_cmq_req *)desc.data;
2253 
2254 	hns_roce_cmq_setup_basic_desc(&desc, HNS_ROCE_OPC_CFG_ENTRY_SIZE,
2255 				      false);
2256 
2257 	hr_reg_write(req, CFG_HEM_ENTRY_SIZE_TYPE, type);
2258 	hr_reg_write(req, CFG_HEM_ENTRY_SIZE_VALUE, val);
2259 
2260 	return hns_roce_cmq_send(hr_dev, &desc, 1);
2261 }
2262 
2263 static int hns_roce_config_entry_size(struct hns_roce_dev *hr_dev)
2264 {
2265 	struct hns_roce_caps *caps = &hr_dev->caps;
2266 	int ret;
2267 
2268 	if (hr_dev->pci_dev->revision == PCI_REVISION_ID_HIP08)
2269 		return 0;
2270 
2271 	ret = config_hem_entry_size(hr_dev, HNS_ROCE_CFG_QPC_SIZE,
2272 				    caps->qpc_sz);
2273 	if (ret) {
2274 		dev_err(hr_dev->dev, "failed to cfg qpc sz, ret = %d.\n", ret);
2275 		return ret;
2276 	}
2277 
2278 	ret = config_hem_entry_size(hr_dev, HNS_ROCE_CFG_SCCC_SIZE,
2279 				    caps->sccc_sz);
2280 	if (ret)
2281 		dev_err(hr_dev->dev, "failed to cfg sccc sz, ret = %d.\n", ret);
2282 
2283 	return ret;
2284 }
2285 
2286 static int hns_roce_v2_vf_profile(struct hns_roce_dev *hr_dev)
2287 {
2288 	struct device *dev = hr_dev->dev;
2289 	int ret;
2290 
2291 	hr_dev->func_num = 1;
2292 
2293 	ret = hns_roce_query_caps(hr_dev);
2294 	if (ret) {
2295 		dev_err(dev, "failed to query VF caps, ret = %d.\n", ret);
2296 		return ret;
2297 	}
2298 
2299 	ret = hns_roce_query_vf_resource(hr_dev);
2300 	if (ret) {
2301 		dev_err(dev, "failed to query VF resource, ret = %d.\n", ret);
2302 		return ret;
2303 	}
2304 
2305 	apply_func_caps(hr_dev);
2306 
2307 	ret = hns_roce_v2_set_bt(hr_dev);
2308 	if (ret)
2309 		dev_err(dev, "failed to config VF BA table, ret = %d.\n", ret);
2310 
2311 	return ret;
2312 }
2313 
2314 static int hns_roce_v2_pf_profile(struct hns_roce_dev *hr_dev)
2315 {
2316 	struct device *dev = hr_dev->dev;
2317 	int ret;
2318 
2319 	ret = hns_roce_query_func_info(hr_dev);
2320 	if (ret) {
2321 		dev_err(dev, "failed to query func info, ret = %d.\n", ret);
2322 		return ret;
2323 	}
2324 
2325 	ret = hns_roce_config_global_param(hr_dev);
2326 	if (ret) {
2327 		dev_err(dev, "failed to config global param, ret = %d.\n", ret);
2328 		return ret;
2329 	}
2330 
2331 	ret = hns_roce_set_vf_switch_param(hr_dev);
2332 	if (ret) {
2333 		dev_err(dev, "failed to set switch param, ret = %d.\n", ret);
2334 		return ret;
2335 	}
2336 
2337 	ret = hns_roce_query_caps(hr_dev);
2338 	if (ret) {
2339 		dev_err(dev, "failed to query PF caps, ret = %d.\n", ret);
2340 		return ret;
2341 	}
2342 
2343 	ret = hns_roce_query_pf_resource(hr_dev);
2344 	if (ret) {
2345 		dev_err(dev, "failed to query pf resource, ret = %d.\n", ret);
2346 		return ret;
2347 	}
2348 
2349 	apply_func_caps(hr_dev);
2350 
2351 	ret = hns_roce_alloc_vf_resource(hr_dev);
2352 	if (ret) {
2353 		dev_err(dev, "failed to alloc vf resource, ret = %d.\n", ret);
2354 		return ret;
2355 	}
2356 
2357 	ret = hns_roce_v2_set_bt(hr_dev);
2358 	if (ret) {
2359 		dev_err(dev, "failed to config BA table, ret = %d.\n", ret);
2360 		return ret;
2361 	}
2362 
2363 	/* Configure the size of QPC, SCCC, etc. */
2364 	return hns_roce_config_entry_size(hr_dev);
2365 }
2366 
2367 static int hns_roce_v2_profile(struct hns_roce_dev *hr_dev)
2368 {
2369 	struct device *dev = hr_dev->dev;
2370 	int ret;
2371 
2372 	ret = hns_roce_cmq_query_hw_info(hr_dev);
2373 	if (ret) {
2374 		dev_err(dev, "failed to query hardware info, ret = %d.\n", ret);
2375 		return ret;
2376 	}
2377 
2378 	ret = hns_roce_query_fw_ver(hr_dev);
2379 	if (ret) {
2380 		dev_err(dev, "failed to query firmware info, ret = %d.\n", ret);
2381 		return ret;
2382 	}
2383 
2384 	hr_dev->vendor_part_id = hr_dev->pci_dev->device;
2385 	hr_dev->sys_image_guid = be64_to_cpu(hr_dev->ib_dev.node_guid);
2386 
2387 	if (hr_dev->is_vf)
2388 		return hns_roce_v2_vf_profile(hr_dev);
2389 	else
2390 		return hns_roce_v2_pf_profile(hr_dev);
2391 }
2392 
2393 static void config_llm_table(struct hns_roce_buf *data_buf, void *cfg_buf)
2394 {
2395 	u32 i, next_ptr, page_num;
2396 	__le64 *entry = cfg_buf;
2397 	dma_addr_t addr;
2398 	u64 val;
2399 
2400 	page_num = data_buf->npages;
2401 	for (i = 0; i < page_num; i++) {
2402 		addr = hns_roce_buf_page(data_buf, i);
2403 		if (i == (page_num - 1))
2404 			next_ptr = 0;
2405 		else
2406 			next_ptr = i + 1;
2407 
2408 		val = HNS_ROCE_EXT_LLM_ENTRY(addr, (u64)next_ptr);
2409 		entry[i] = cpu_to_le64(val);
2410 	}
2411 }
2412 
2413 static int set_llm_cfg_to_hw(struct hns_roce_dev *hr_dev,
2414 			     struct hns_roce_link_table *table)
2415 {
2416 	struct hns_roce_cmq_desc desc[2];
2417 	struct hns_roce_cmq_req *r_a = (struct hns_roce_cmq_req *)desc[0].data;
2418 	struct hns_roce_cmq_req *r_b = (struct hns_roce_cmq_req *)desc[1].data;
2419 	struct hns_roce_buf *buf = table->buf;
2420 	enum hns_roce_opcode_type opcode;
2421 	dma_addr_t addr;
2422 
2423 	opcode = HNS_ROCE_OPC_CFG_EXT_LLM;
2424 	hns_roce_cmq_setup_basic_desc(&desc[0], opcode, false);
2425 	desc[0].flag |= cpu_to_le16(HNS_ROCE_CMD_FLAG_NEXT);
2426 	hns_roce_cmq_setup_basic_desc(&desc[1], opcode, false);
2427 
2428 	hr_reg_write(r_a, CFG_LLM_A_BA_L, lower_32_bits(table->table.map));
2429 	hr_reg_write(r_a, CFG_LLM_A_BA_H, upper_32_bits(table->table.map));
2430 	hr_reg_write(r_a, CFG_LLM_A_DEPTH, buf->npages);
2431 	hr_reg_write(r_a, CFG_LLM_A_PGSZ, to_hr_hw_page_shift(buf->page_shift));
2432 	hr_reg_enable(r_a, CFG_LLM_A_INIT_EN);
2433 
2434 	addr = to_hr_hw_page_addr(hns_roce_buf_page(buf, 0));
2435 	hr_reg_write(r_a, CFG_LLM_A_HEAD_BA_L, lower_32_bits(addr));
2436 	hr_reg_write(r_a, CFG_LLM_A_HEAD_BA_H, upper_32_bits(addr));
2437 	hr_reg_write(r_a, CFG_LLM_A_HEAD_NXTPTR, 1);
2438 	hr_reg_write(r_a, CFG_LLM_A_HEAD_PTR, 0);
2439 
2440 	addr = to_hr_hw_page_addr(hns_roce_buf_page(buf, buf->npages - 1));
2441 	hr_reg_write(r_b, CFG_LLM_B_TAIL_BA_L, lower_32_bits(addr));
2442 	hr_reg_write(r_b, CFG_LLM_B_TAIL_BA_H, upper_32_bits(addr));
2443 	hr_reg_write(r_b, CFG_LLM_B_TAIL_PTR, buf->npages - 1);
2444 
2445 	return hns_roce_cmq_send(hr_dev, desc, 2);
2446 }
2447 
2448 static struct hns_roce_link_table *
2449 alloc_link_table_buf(struct hns_roce_dev *hr_dev)
2450 {
2451 	u16 total_sl = hr_dev->caps.sl_num * hr_dev->func_num;
2452 	struct hns_roce_v2_priv *priv = hr_dev->priv;
2453 	struct hns_roce_link_table *link_tbl;
2454 	u32 pg_shift, size, min_size;
2455 
2456 	link_tbl = &priv->ext_llm;
2457 	pg_shift = hr_dev->caps.llm_buf_pg_sz + PAGE_SHIFT;
2458 	size = hr_dev->caps.num_qps * hr_dev->func_num *
2459 	       HNS_ROCE_V2_EXT_LLM_ENTRY_SZ;
2460 	min_size = HNS_ROCE_EXT_LLM_MIN_PAGES(total_sl) << pg_shift;
2461 
2462 	/* Alloc data table */
2463 	size = max(size, min_size);
2464 	link_tbl->buf = hns_roce_buf_alloc(hr_dev, size, pg_shift, 0);
2465 	if (IS_ERR(link_tbl->buf))
2466 		return ERR_PTR(-ENOMEM);
2467 
2468 	/* Alloc config table */
2469 	size = link_tbl->buf->npages * sizeof(u64);
2470 	link_tbl->table.buf = dma_alloc_coherent(hr_dev->dev, size,
2471 						 &link_tbl->table.map,
2472 						 GFP_KERNEL);
2473 	if (!link_tbl->table.buf) {
2474 		hns_roce_buf_free(hr_dev, link_tbl->buf);
2475 		return ERR_PTR(-ENOMEM);
2476 	}
2477 
2478 	return link_tbl;
2479 }
2480 
2481 static void free_link_table_buf(struct hns_roce_dev *hr_dev,
2482 				struct hns_roce_link_table *tbl)
2483 {
2484 	if (tbl->buf) {
2485 		u32 size = tbl->buf->npages * sizeof(u64);
2486 
2487 		dma_free_coherent(hr_dev->dev, size, tbl->table.buf,
2488 				  tbl->table.map);
2489 	}
2490 
2491 	hns_roce_buf_free(hr_dev, tbl->buf);
2492 }
2493 
2494 static int hns_roce_init_link_table(struct hns_roce_dev *hr_dev)
2495 {
2496 	struct hns_roce_link_table *link_tbl;
2497 	int ret;
2498 
2499 	link_tbl = alloc_link_table_buf(hr_dev);
2500 	if (IS_ERR(link_tbl))
2501 		return -ENOMEM;
2502 
2503 	if (WARN_ON(link_tbl->buf->npages > HNS_ROCE_V2_EXT_LLM_MAX_DEPTH)) {
2504 		ret = -EINVAL;
2505 		goto err_alloc;
2506 	}
2507 
2508 	config_llm_table(link_tbl->buf, link_tbl->table.buf);
2509 	ret = set_llm_cfg_to_hw(hr_dev, link_tbl);
2510 	if (ret)
2511 		goto err_alloc;
2512 
2513 	return 0;
2514 
2515 err_alloc:
2516 	free_link_table_buf(hr_dev, link_tbl);
2517 	return ret;
2518 }
2519 
2520 static void hns_roce_free_link_table(struct hns_roce_dev *hr_dev)
2521 {
2522 	struct hns_roce_v2_priv *priv = hr_dev->priv;
2523 
2524 	free_link_table_buf(hr_dev, &priv->ext_llm);
2525 }
2526 
2527 static void free_dip_list(struct hns_roce_dev *hr_dev)
2528 {
2529 	struct hns_roce_dip *hr_dip;
2530 	struct hns_roce_dip *tmp;
2531 	unsigned long flags;
2532 
2533 	spin_lock_irqsave(&hr_dev->dip_list_lock, flags);
2534 
2535 	list_for_each_entry_safe(hr_dip, tmp, &hr_dev->dip_list, node) {
2536 		list_del(&hr_dip->node);
2537 		kfree(hr_dip);
2538 	}
2539 
2540 	spin_unlock_irqrestore(&hr_dev->dip_list_lock, flags);
2541 }
2542 
2543 static struct ib_pd *free_mr_init_pd(struct hns_roce_dev *hr_dev)
2544 {
2545 	struct hns_roce_v2_priv *priv = hr_dev->priv;
2546 	struct hns_roce_v2_free_mr *free_mr = &priv->free_mr;
2547 	struct ib_device *ibdev = &hr_dev->ib_dev;
2548 	struct hns_roce_pd *hr_pd;
2549 	struct ib_pd *pd;
2550 
2551 	hr_pd = kzalloc(sizeof(*hr_pd), GFP_KERNEL);
2552 	if (ZERO_OR_NULL_PTR(hr_pd))
2553 		return NULL;
2554 	pd = &hr_pd->ibpd;
2555 	pd->device = ibdev;
2556 
2557 	if (hns_roce_alloc_pd(pd, NULL)) {
2558 		ibdev_err(ibdev, "failed to create pd for free mr.\n");
2559 		kfree(hr_pd);
2560 		return NULL;
2561 	}
2562 	free_mr->rsv_pd = to_hr_pd(pd);
2563 	free_mr->rsv_pd->ibpd.device = &hr_dev->ib_dev;
2564 	free_mr->rsv_pd->ibpd.uobject = NULL;
2565 	free_mr->rsv_pd->ibpd.__internal_mr = NULL;
2566 	atomic_set(&free_mr->rsv_pd->ibpd.usecnt, 0);
2567 
2568 	return pd;
2569 }
2570 
2571 static struct ib_cq *free_mr_init_cq(struct hns_roce_dev *hr_dev)
2572 {
2573 	struct hns_roce_v2_priv *priv = hr_dev->priv;
2574 	struct hns_roce_v2_free_mr *free_mr = &priv->free_mr;
2575 	struct ib_device *ibdev = &hr_dev->ib_dev;
2576 	struct ib_cq_init_attr cq_init_attr = {};
2577 	struct hns_roce_cq *hr_cq;
2578 	struct ib_cq *cq;
2579 
2580 	cq_init_attr.cqe = HNS_ROCE_FREE_MR_USED_CQE_NUM;
2581 
2582 	hr_cq = kzalloc(sizeof(*hr_cq), GFP_KERNEL);
2583 	if (ZERO_OR_NULL_PTR(hr_cq))
2584 		return NULL;
2585 
2586 	cq = &hr_cq->ib_cq;
2587 	cq->device = ibdev;
2588 
2589 	if (hns_roce_create_cq(cq, &cq_init_attr, NULL)) {
2590 		ibdev_err(ibdev, "failed to create cq for free mr.\n");
2591 		kfree(hr_cq);
2592 		return NULL;
2593 	}
2594 	free_mr->rsv_cq = to_hr_cq(cq);
2595 	free_mr->rsv_cq->ib_cq.device = &hr_dev->ib_dev;
2596 	free_mr->rsv_cq->ib_cq.uobject = NULL;
2597 	free_mr->rsv_cq->ib_cq.comp_handler = NULL;
2598 	free_mr->rsv_cq->ib_cq.event_handler = NULL;
2599 	free_mr->rsv_cq->ib_cq.cq_context = NULL;
2600 	atomic_set(&free_mr->rsv_cq->ib_cq.usecnt, 0);
2601 
2602 	return cq;
2603 }
2604 
2605 static int free_mr_init_qp(struct hns_roce_dev *hr_dev, struct ib_cq *cq,
2606 			   struct ib_qp_init_attr *init_attr, int i)
2607 {
2608 	struct hns_roce_v2_priv *priv = hr_dev->priv;
2609 	struct hns_roce_v2_free_mr *free_mr = &priv->free_mr;
2610 	struct ib_device *ibdev = &hr_dev->ib_dev;
2611 	struct hns_roce_qp *hr_qp;
2612 	struct ib_qp *qp;
2613 	int ret;
2614 
2615 	hr_qp = kzalloc(sizeof(*hr_qp), GFP_KERNEL);
2616 	if (ZERO_OR_NULL_PTR(hr_qp))
2617 		return -ENOMEM;
2618 
2619 	qp = &hr_qp->ibqp;
2620 	qp->device = ibdev;
2621 
2622 	ret = hns_roce_create_qp(qp, init_attr, NULL);
2623 	if (ret) {
2624 		ibdev_err(ibdev, "failed to create qp for free mr.\n");
2625 		kfree(hr_qp);
2626 		return ret;
2627 	}
2628 
2629 	free_mr->rsv_qp[i] = hr_qp;
2630 	free_mr->rsv_qp[i]->ibqp.recv_cq = cq;
2631 	free_mr->rsv_qp[i]->ibqp.send_cq = cq;
2632 
2633 	return 0;
2634 }
2635 
2636 static void free_mr_exit(struct hns_roce_dev *hr_dev)
2637 {
2638 	struct hns_roce_v2_priv *priv = hr_dev->priv;
2639 	struct hns_roce_v2_free_mr *free_mr = &priv->free_mr;
2640 	struct ib_qp *qp;
2641 	int i;
2642 
2643 	for (i = 0; i < ARRAY_SIZE(free_mr->rsv_qp); i++) {
2644 		if (free_mr->rsv_qp[i]) {
2645 			qp = &free_mr->rsv_qp[i]->ibqp;
2646 			hns_roce_v2_destroy_qp(qp, NULL);
2647 			kfree(free_mr->rsv_qp[i]);
2648 			free_mr->rsv_qp[i] = NULL;
2649 		}
2650 	}
2651 
2652 	if (free_mr->rsv_cq) {
2653 		hns_roce_destroy_cq(&free_mr->rsv_cq->ib_cq, NULL);
2654 		kfree(free_mr->rsv_cq);
2655 		free_mr->rsv_cq = NULL;
2656 	}
2657 
2658 	if (free_mr->rsv_pd) {
2659 		hns_roce_dealloc_pd(&free_mr->rsv_pd->ibpd, NULL);
2660 		kfree(free_mr->rsv_pd);
2661 		free_mr->rsv_pd = NULL;
2662 	}
2663 }
2664 
2665 static int free_mr_alloc_res(struct hns_roce_dev *hr_dev)
2666 {
2667 	struct hns_roce_v2_priv *priv = hr_dev->priv;
2668 	struct hns_roce_v2_free_mr *free_mr = &priv->free_mr;
2669 	struct ib_qp_init_attr qp_init_attr = {};
2670 	struct ib_pd *pd;
2671 	struct ib_cq *cq;
2672 	int ret;
2673 	int i;
2674 
2675 	pd = free_mr_init_pd(hr_dev);
2676 	if (!pd)
2677 		return -ENOMEM;
2678 
2679 	cq = free_mr_init_cq(hr_dev);
2680 	if (!cq) {
2681 		ret = -ENOMEM;
2682 		goto create_failed_cq;
2683 	}
2684 
2685 	qp_init_attr.qp_type = IB_QPT_RC;
2686 	qp_init_attr.sq_sig_type = IB_SIGNAL_ALL_WR;
2687 	qp_init_attr.send_cq = cq;
2688 	qp_init_attr.recv_cq = cq;
2689 	for (i = 0; i < ARRAY_SIZE(free_mr->rsv_qp); i++) {
2690 		qp_init_attr.cap.max_send_wr = HNS_ROCE_FREE_MR_USED_SQWQE_NUM;
2691 		qp_init_attr.cap.max_send_sge = HNS_ROCE_FREE_MR_USED_SQSGE_NUM;
2692 		qp_init_attr.cap.max_recv_wr = HNS_ROCE_FREE_MR_USED_RQWQE_NUM;
2693 		qp_init_attr.cap.max_recv_sge = HNS_ROCE_FREE_MR_USED_RQSGE_NUM;
2694 
2695 		ret = free_mr_init_qp(hr_dev, cq, &qp_init_attr, i);
2696 		if (ret)
2697 			goto create_failed_qp;
2698 	}
2699 
2700 	return 0;
2701 
2702 create_failed_qp:
2703 	for (i--; i >= 0; i--) {
2704 		hns_roce_v2_destroy_qp(&free_mr->rsv_qp[i]->ibqp, NULL);
2705 		kfree(free_mr->rsv_qp[i]);
2706 	}
2707 	hns_roce_destroy_cq(cq, NULL);
2708 	kfree(cq);
2709 
2710 create_failed_cq:
2711 	hns_roce_dealloc_pd(pd, NULL);
2712 	kfree(pd);
2713 
2714 	return ret;
2715 }
2716 
2717 static int free_mr_modify_rsv_qp(struct hns_roce_dev *hr_dev,
2718 				 struct ib_qp_attr *attr, int sl_num)
2719 {
2720 	struct hns_roce_v2_priv *priv = hr_dev->priv;
2721 	struct hns_roce_v2_free_mr *free_mr = &priv->free_mr;
2722 	struct ib_device *ibdev = &hr_dev->ib_dev;
2723 	struct hns_roce_qp *hr_qp;
2724 	int loopback;
2725 	int mask;
2726 	int ret;
2727 
2728 	hr_qp = to_hr_qp(&free_mr->rsv_qp[sl_num]->ibqp);
2729 	hr_qp->free_mr_en = 1;
2730 	hr_qp->ibqp.device = ibdev;
2731 	hr_qp->ibqp.qp_type = IB_QPT_RC;
2732 
2733 	mask = IB_QP_STATE | IB_QP_PKEY_INDEX | IB_QP_PORT | IB_QP_ACCESS_FLAGS;
2734 	attr->qp_state = IB_QPS_INIT;
2735 	attr->port_num = 1;
2736 	attr->qp_access_flags = IB_ACCESS_REMOTE_WRITE;
2737 	ret = hr_dev->hw->modify_qp(&hr_qp->ibqp, attr, mask, IB_QPS_INIT,
2738 				    IB_QPS_INIT, NULL);
2739 	if (ret) {
2740 		ibdev_err(ibdev, "failed to modify qp to init, ret = %d.\n",
2741 			  ret);
2742 		return ret;
2743 	}
2744 
2745 	loopback = hr_dev->loop_idc;
2746 	/* Set qpc lbi = 1 incidate loopback IO */
2747 	hr_dev->loop_idc = 1;
2748 
2749 	mask = IB_QP_STATE | IB_QP_AV | IB_QP_PATH_MTU | IB_QP_DEST_QPN |
2750 	       IB_QP_RQ_PSN | IB_QP_MAX_DEST_RD_ATOMIC | IB_QP_MIN_RNR_TIMER;
2751 	attr->qp_state = IB_QPS_RTR;
2752 	attr->ah_attr.type = RDMA_AH_ATTR_TYPE_ROCE;
2753 	attr->path_mtu = IB_MTU_256;
2754 	attr->dest_qp_num = hr_qp->qpn;
2755 	attr->rq_psn = HNS_ROCE_FREE_MR_USED_PSN;
2756 
2757 	rdma_ah_set_sl(&attr->ah_attr, (u8)sl_num);
2758 
2759 	ret = hr_dev->hw->modify_qp(&hr_qp->ibqp, attr, mask, IB_QPS_INIT,
2760 				    IB_QPS_RTR, NULL);
2761 	hr_dev->loop_idc = loopback;
2762 	if (ret) {
2763 		ibdev_err(ibdev, "failed to modify qp to rtr, ret = %d.\n",
2764 			  ret);
2765 		return ret;
2766 	}
2767 
2768 	mask = IB_QP_STATE | IB_QP_SQ_PSN | IB_QP_RETRY_CNT | IB_QP_TIMEOUT |
2769 	       IB_QP_RNR_RETRY | IB_QP_MAX_QP_RD_ATOMIC;
2770 	attr->qp_state = IB_QPS_RTS;
2771 	attr->sq_psn = HNS_ROCE_FREE_MR_USED_PSN;
2772 	attr->retry_cnt = HNS_ROCE_FREE_MR_USED_QP_RETRY_CNT;
2773 	attr->timeout = HNS_ROCE_FREE_MR_USED_QP_TIMEOUT;
2774 	ret = hr_dev->hw->modify_qp(&hr_qp->ibqp, attr, mask, IB_QPS_RTR,
2775 				    IB_QPS_RTS, NULL);
2776 	if (ret)
2777 		ibdev_err(ibdev, "failed to modify qp to rts, ret = %d.\n",
2778 			  ret);
2779 
2780 	return ret;
2781 }
2782 
2783 static int free_mr_modify_qp(struct hns_roce_dev *hr_dev)
2784 {
2785 	struct hns_roce_v2_priv *priv = hr_dev->priv;
2786 	struct hns_roce_v2_free_mr *free_mr = &priv->free_mr;
2787 	struct ib_qp_attr attr = {};
2788 	int ret;
2789 	int i;
2790 
2791 	rdma_ah_set_grh(&attr.ah_attr, NULL, 0, 0, 1, 0);
2792 	rdma_ah_set_static_rate(&attr.ah_attr, 3);
2793 	rdma_ah_set_port_num(&attr.ah_attr, 1);
2794 
2795 	for (i = 0; i < ARRAY_SIZE(free_mr->rsv_qp); i++) {
2796 		ret = free_mr_modify_rsv_qp(hr_dev, &attr, i);
2797 		if (ret)
2798 			return ret;
2799 	}
2800 
2801 	return 0;
2802 }
2803 
2804 static int free_mr_init(struct hns_roce_dev *hr_dev)
2805 {
2806 	struct hns_roce_v2_priv *priv = hr_dev->priv;
2807 	struct hns_roce_v2_free_mr *free_mr = &priv->free_mr;
2808 	int ret;
2809 
2810 	mutex_init(&free_mr->mutex);
2811 
2812 	ret = free_mr_alloc_res(hr_dev);
2813 	if (ret)
2814 		return ret;
2815 
2816 	ret = free_mr_modify_qp(hr_dev);
2817 	if (ret)
2818 		goto err_modify_qp;
2819 
2820 	return 0;
2821 
2822 err_modify_qp:
2823 	free_mr_exit(hr_dev);
2824 
2825 	return ret;
2826 }
2827 
2828 static int get_hem_table(struct hns_roce_dev *hr_dev)
2829 {
2830 	unsigned int qpc_count;
2831 	unsigned int cqc_count;
2832 	unsigned int gmv_count;
2833 	int ret;
2834 	int i;
2835 
2836 	/* Alloc memory for source address table buffer space chunk */
2837 	for (gmv_count = 0; gmv_count < hr_dev->caps.gmv_entry_num;
2838 	     gmv_count++) {
2839 		ret = hns_roce_table_get(hr_dev, &hr_dev->gmv_table, gmv_count);
2840 		if (ret)
2841 			goto err_gmv_failed;
2842 	}
2843 
2844 	if (hr_dev->is_vf)
2845 		return 0;
2846 
2847 	/* Alloc memory for QPC Timer buffer space chunk */
2848 	for (qpc_count = 0; qpc_count < hr_dev->caps.qpc_timer_bt_num;
2849 	     qpc_count++) {
2850 		ret = hns_roce_table_get(hr_dev, &hr_dev->qpc_timer_table,
2851 					 qpc_count);
2852 		if (ret) {
2853 			dev_err(hr_dev->dev, "QPC Timer get failed\n");
2854 			goto err_qpc_timer_failed;
2855 		}
2856 	}
2857 
2858 	/* Alloc memory for CQC Timer buffer space chunk */
2859 	for (cqc_count = 0; cqc_count < hr_dev->caps.cqc_timer_bt_num;
2860 	     cqc_count++) {
2861 		ret = hns_roce_table_get(hr_dev, &hr_dev->cqc_timer_table,
2862 					 cqc_count);
2863 		if (ret) {
2864 			dev_err(hr_dev->dev, "CQC Timer get failed\n");
2865 			goto err_cqc_timer_failed;
2866 		}
2867 	}
2868 
2869 	return 0;
2870 
2871 err_cqc_timer_failed:
2872 	for (i = 0; i < cqc_count; i++)
2873 		hns_roce_table_put(hr_dev, &hr_dev->cqc_timer_table, i);
2874 
2875 err_qpc_timer_failed:
2876 	for (i = 0; i < qpc_count; i++)
2877 		hns_roce_table_put(hr_dev, &hr_dev->qpc_timer_table, i);
2878 
2879 err_gmv_failed:
2880 	for (i = 0; i < gmv_count; i++)
2881 		hns_roce_table_put(hr_dev, &hr_dev->gmv_table, i);
2882 
2883 	return ret;
2884 }
2885 
2886 static void put_hem_table(struct hns_roce_dev *hr_dev)
2887 {
2888 	int i;
2889 
2890 	for (i = 0; i < hr_dev->caps.gmv_entry_num; i++)
2891 		hns_roce_table_put(hr_dev, &hr_dev->gmv_table, i);
2892 
2893 	if (hr_dev->is_vf)
2894 		return;
2895 
2896 	for (i = 0; i < hr_dev->caps.qpc_timer_bt_num; i++)
2897 		hns_roce_table_put(hr_dev, &hr_dev->qpc_timer_table, i);
2898 
2899 	for (i = 0; i < hr_dev->caps.cqc_timer_bt_num; i++)
2900 		hns_roce_table_put(hr_dev, &hr_dev->cqc_timer_table, i);
2901 }
2902 
2903 static int hns_roce_v2_init(struct hns_roce_dev *hr_dev)
2904 {
2905 	int ret;
2906 
2907 	/* The hns ROCEE requires the extdb info to be cleared before using */
2908 	ret = hns_roce_clear_extdb_list_info(hr_dev);
2909 	if (ret)
2910 		return ret;
2911 
2912 	ret = get_hem_table(hr_dev);
2913 	if (ret)
2914 		return ret;
2915 
2916 	if (hr_dev->is_vf)
2917 		return 0;
2918 
2919 	ret = hns_roce_init_link_table(hr_dev);
2920 	if (ret) {
2921 		dev_err(hr_dev->dev, "failed to init llm, ret = %d.\n", ret);
2922 		goto err_llm_init_failed;
2923 	}
2924 
2925 	return 0;
2926 
2927 err_llm_init_failed:
2928 	put_hem_table(hr_dev);
2929 
2930 	return ret;
2931 }
2932 
2933 static void hns_roce_v2_exit(struct hns_roce_dev *hr_dev)
2934 {
2935 	if (hr_dev->pci_dev->revision == PCI_REVISION_ID_HIP08)
2936 		free_mr_exit(hr_dev);
2937 
2938 	hns_roce_function_clear(hr_dev);
2939 
2940 	if (!hr_dev->is_vf)
2941 		hns_roce_free_link_table(hr_dev);
2942 
2943 	if (hr_dev->pci_dev->revision == PCI_REVISION_ID_HIP09)
2944 		free_dip_list(hr_dev);
2945 }
2946 
2947 static int hns_roce_mbox_post(struct hns_roce_dev *hr_dev,
2948 			      struct hns_roce_mbox_msg *mbox_msg)
2949 {
2950 	struct hns_roce_cmq_desc desc;
2951 	struct hns_roce_post_mbox *mb = (struct hns_roce_post_mbox *)desc.data;
2952 
2953 	hns_roce_cmq_setup_basic_desc(&desc, HNS_ROCE_OPC_POST_MB, false);
2954 
2955 	mb->in_param_l = cpu_to_le32(mbox_msg->in_param);
2956 	mb->in_param_h = cpu_to_le32(mbox_msg->in_param >> 32);
2957 	mb->out_param_l = cpu_to_le32(mbox_msg->out_param);
2958 	mb->out_param_h = cpu_to_le32(mbox_msg->out_param >> 32);
2959 	mb->cmd_tag = cpu_to_le32(mbox_msg->tag << 8 | mbox_msg->cmd);
2960 	mb->token_event_en = cpu_to_le32(mbox_msg->event_en << 16 |
2961 					 mbox_msg->token);
2962 
2963 	return hns_roce_cmq_send(hr_dev, &desc, 1);
2964 }
2965 
2966 static int v2_wait_mbox_complete(struct hns_roce_dev *hr_dev, u32 timeout,
2967 				 u8 *complete_status)
2968 {
2969 	struct hns_roce_mbox_status *mb_st;
2970 	struct hns_roce_cmq_desc desc;
2971 	unsigned long end;
2972 	int ret = -EBUSY;
2973 	u32 status;
2974 	bool busy;
2975 
2976 	mb_st = (struct hns_roce_mbox_status *)desc.data;
2977 	end = msecs_to_jiffies(timeout) + jiffies;
2978 	while (v2_chk_mbox_is_avail(hr_dev, &busy)) {
2979 		if (hr_dev->cmd.state == HNS_ROCE_CMDQ_STATE_FATAL_ERR)
2980 			return -EIO;
2981 
2982 		status = 0;
2983 		hns_roce_cmq_setup_basic_desc(&desc, HNS_ROCE_OPC_QUERY_MB_ST,
2984 					      true);
2985 		ret = __hns_roce_cmq_send(hr_dev, &desc, 1);
2986 		if (!ret) {
2987 			status = le32_to_cpu(mb_st->mb_status_hw_run);
2988 			/* No pending message exists in ROCEE mbox. */
2989 			if (!(status & MB_ST_HW_RUN_M))
2990 				break;
2991 		} else if (!v2_chk_mbox_is_avail(hr_dev, &busy)) {
2992 			break;
2993 		}
2994 
2995 		if (time_after(jiffies, end)) {
2996 			dev_err_ratelimited(hr_dev->dev,
2997 					    "failed to wait mbox status 0x%x\n",
2998 					    status);
2999 			return -ETIMEDOUT;
3000 		}
3001 
3002 		cond_resched();
3003 		ret = -EBUSY;
3004 	}
3005 
3006 	if (!ret) {
3007 		*complete_status = (u8)(status & MB_ST_COMPLETE_M);
3008 	} else if (!v2_chk_mbox_is_avail(hr_dev, &busy)) {
3009 		/* Ignore all errors if the mbox is unavailable. */
3010 		ret = 0;
3011 		*complete_status = MB_ST_COMPLETE_M;
3012 	}
3013 
3014 	return ret;
3015 }
3016 
3017 static int v2_post_mbox(struct hns_roce_dev *hr_dev,
3018 			struct hns_roce_mbox_msg *mbox_msg)
3019 {
3020 	u8 status = 0;
3021 	int ret;
3022 
3023 	/* Waiting for the mbox to be idle */
3024 	ret = v2_wait_mbox_complete(hr_dev, HNS_ROCE_V2_GO_BIT_TIMEOUT_MSECS,
3025 				    &status);
3026 	if (unlikely(ret)) {
3027 		dev_err_ratelimited(hr_dev->dev,
3028 				    "failed to check post mbox status = 0x%x, ret = %d.\n",
3029 				    status, ret);
3030 		return ret;
3031 	}
3032 
3033 	/* Post new message to mbox */
3034 	ret = hns_roce_mbox_post(hr_dev, mbox_msg);
3035 	if (ret)
3036 		dev_err_ratelimited(hr_dev->dev,
3037 				    "failed to post mailbox, ret = %d.\n", ret);
3038 
3039 	return ret;
3040 }
3041 
3042 static int v2_poll_mbox_done(struct hns_roce_dev *hr_dev)
3043 {
3044 	u8 status = 0;
3045 	int ret;
3046 
3047 	ret = v2_wait_mbox_complete(hr_dev, HNS_ROCE_CMD_TIMEOUT_MSECS,
3048 				    &status);
3049 	if (!ret) {
3050 		if (status != MB_ST_COMPLETE_SUCC)
3051 			return -EBUSY;
3052 	} else {
3053 		dev_err_ratelimited(hr_dev->dev,
3054 				    "failed to check mbox status = 0x%x, ret = %d.\n",
3055 				    status, ret);
3056 	}
3057 
3058 	return ret;
3059 }
3060 
3061 static void copy_gid(void *dest, const union ib_gid *gid)
3062 {
3063 #define GID_SIZE 4
3064 	const union ib_gid *src = gid;
3065 	__le32 (*p)[GID_SIZE] = dest;
3066 	int i;
3067 
3068 	if (!gid)
3069 		src = &zgid;
3070 
3071 	for (i = 0; i < GID_SIZE; i++)
3072 		(*p)[i] = cpu_to_le32(*(u32 *)&src->raw[i * sizeof(u32)]);
3073 }
3074 
3075 static int config_sgid_table(struct hns_roce_dev *hr_dev,
3076 			     int gid_index, const union ib_gid *gid,
3077 			     enum hns_roce_sgid_type sgid_type)
3078 {
3079 	struct hns_roce_cmq_desc desc;
3080 	struct hns_roce_cfg_sgid_tb *sgid_tb =
3081 				    (struct hns_roce_cfg_sgid_tb *)desc.data;
3082 
3083 	hns_roce_cmq_setup_basic_desc(&desc, HNS_ROCE_OPC_CFG_SGID_TB, false);
3084 
3085 	hr_reg_write(sgid_tb, CFG_SGID_TB_TABLE_IDX, gid_index);
3086 	hr_reg_write(sgid_tb, CFG_SGID_TB_VF_SGID_TYPE, sgid_type);
3087 
3088 	copy_gid(&sgid_tb->vf_sgid_l, gid);
3089 
3090 	return hns_roce_cmq_send(hr_dev, &desc, 1);
3091 }
3092 
3093 static int config_gmv_table(struct hns_roce_dev *hr_dev,
3094 			    int gid_index, const union ib_gid *gid,
3095 			    enum hns_roce_sgid_type sgid_type,
3096 			    const struct ib_gid_attr *attr)
3097 {
3098 	struct hns_roce_cmq_desc desc[2];
3099 	struct hns_roce_cfg_gmv_tb_a *tb_a =
3100 				(struct hns_roce_cfg_gmv_tb_a *)desc[0].data;
3101 	struct hns_roce_cfg_gmv_tb_b *tb_b =
3102 				(struct hns_roce_cfg_gmv_tb_b *)desc[1].data;
3103 
3104 	u16 vlan_id = VLAN_CFI_MASK;
3105 	u8 mac[ETH_ALEN] = {};
3106 	int ret;
3107 
3108 	if (gid) {
3109 		ret = rdma_read_gid_l2_fields(attr, &vlan_id, mac);
3110 		if (ret)
3111 			return ret;
3112 	}
3113 
3114 	hns_roce_cmq_setup_basic_desc(&desc[0], HNS_ROCE_OPC_CFG_GMV_TBL, false);
3115 	desc[0].flag |= cpu_to_le16(HNS_ROCE_CMD_FLAG_NEXT);
3116 
3117 	hns_roce_cmq_setup_basic_desc(&desc[1], HNS_ROCE_OPC_CFG_GMV_TBL, false);
3118 
3119 	copy_gid(&tb_a->vf_sgid_l, gid);
3120 
3121 	hr_reg_write(tb_a, GMV_TB_A_VF_SGID_TYPE, sgid_type);
3122 	hr_reg_write(tb_a, GMV_TB_A_VF_VLAN_EN, vlan_id < VLAN_CFI_MASK);
3123 	hr_reg_write(tb_a, GMV_TB_A_VF_VLAN_ID, vlan_id);
3124 
3125 	tb_b->vf_smac_l = cpu_to_le32(*(u32 *)mac);
3126 
3127 	hr_reg_write(tb_b, GMV_TB_B_SMAC_H, *(u16 *)&mac[4]);
3128 	hr_reg_write(tb_b, GMV_TB_B_SGID_IDX, gid_index);
3129 
3130 	return hns_roce_cmq_send(hr_dev, desc, 2);
3131 }
3132 
3133 static int hns_roce_v2_set_gid(struct hns_roce_dev *hr_dev, int gid_index,
3134 			       const union ib_gid *gid,
3135 			       const struct ib_gid_attr *attr)
3136 {
3137 	enum hns_roce_sgid_type sgid_type = GID_TYPE_FLAG_ROCE_V1;
3138 	int ret;
3139 
3140 	if (gid) {
3141 		if (attr->gid_type == IB_GID_TYPE_ROCE_UDP_ENCAP) {
3142 			if (ipv6_addr_v4mapped((void *)gid))
3143 				sgid_type = GID_TYPE_FLAG_ROCE_V2_IPV4;
3144 			else
3145 				sgid_type = GID_TYPE_FLAG_ROCE_V2_IPV6;
3146 		} else if (attr->gid_type == IB_GID_TYPE_ROCE) {
3147 			sgid_type = GID_TYPE_FLAG_ROCE_V1;
3148 		}
3149 	}
3150 
3151 	if (hr_dev->pci_dev->revision >= PCI_REVISION_ID_HIP09)
3152 		ret = config_gmv_table(hr_dev, gid_index, gid, sgid_type, attr);
3153 	else
3154 		ret = config_sgid_table(hr_dev, gid_index, gid, sgid_type);
3155 
3156 	if (ret)
3157 		ibdev_err(&hr_dev->ib_dev, "failed to set gid, ret = %d!\n",
3158 			  ret);
3159 
3160 	return ret;
3161 }
3162 
3163 static int hns_roce_v2_set_mac(struct hns_roce_dev *hr_dev, u8 phy_port,
3164 			       const u8 *addr)
3165 {
3166 	struct hns_roce_cmq_desc desc;
3167 	struct hns_roce_cfg_smac_tb *smac_tb =
3168 				    (struct hns_roce_cfg_smac_tb *)desc.data;
3169 	u16 reg_smac_h;
3170 	u32 reg_smac_l;
3171 
3172 	hns_roce_cmq_setup_basic_desc(&desc, HNS_ROCE_OPC_CFG_SMAC_TB, false);
3173 
3174 	reg_smac_l = *(u32 *)(&addr[0]);
3175 	reg_smac_h = *(u16 *)(&addr[4]);
3176 
3177 	hr_reg_write(smac_tb, CFG_SMAC_TB_IDX, phy_port);
3178 	hr_reg_write(smac_tb, CFG_SMAC_TB_VF_SMAC_H, reg_smac_h);
3179 	smac_tb->vf_smac_l = cpu_to_le32(reg_smac_l);
3180 
3181 	return hns_roce_cmq_send(hr_dev, &desc, 1);
3182 }
3183 
3184 static int set_mtpt_pbl(struct hns_roce_dev *hr_dev,
3185 			struct hns_roce_v2_mpt_entry *mpt_entry,
3186 			struct hns_roce_mr *mr)
3187 {
3188 	u64 pages[HNS_ROCE_V2_MAX_INNER_MTPT_NUM] = { 0 };
3189 	struct ib_device *ibdev = &hr_dev->ib_dev;
3190 	dma_addr_t pbl_ba;
3191 	int i, count;
3192 
3193 	count = hns_roce_mtr_find(hr_dev, &mr->pbl_mtr, 0, pages,
3194 				  min_t(int, ARRAY_SIZE(pages), mr->npages),
3195 				  &pbl_ba);
3196 	if (count < 1) {
3197 		ibdev_err(ibdev, "failed to find PBL mtr, count = %d.\n",
3198 			  count);
3199 		return -ENOBUFS;
3200 	}
3201 
3202 	/* Aligned to the hardware address access unit */
3203 	for (i = 0; i < count; i++)
3204 		pages[i] >>= 6;
3205 
3206 	mpt_entry->pbl_size = cpu_to_le32(mr->npages);
3207 	mpt_entry->pbl_ba_l = cpu_to_le32(pbl_ba >> 3);
3208 	hr_reg_write(mpt_entry, MPT_PBL_BA_H, upper_32_bits(pbl_ba >> 3));
3209 
3210 	mpt_entry->pa0_l = cpu_to_le32(lower_32_bits(pages[0]));
3211 	hr_reg_write(mpt_entry, MPT_PA0_H, upper_32_bits(pages[0]));
3212 
3213 	mpt_entry->pa1_l = cpu_to_le32(lower_32_bits(pages[1]));
3214 	hr_reg_write(mpt_entry, MPT_PA1_H, upper_32_bits(pages[1]));
3215 	hr_reg_write(mpt_entry, MPT_PBL_BUF_PG_SZ,
3216 		     to_hr_hw_page_shift(mr->pbl_mtr.hem_cfg.buf_pg_shift));
3217 
3218 	return 0;
3219 }
3220 
3221 static int hns_roce_v2_write_mtpt(struct hns_roce_dev *hr_dev,
3222 				  void *mb_buf, struct hns_roce_mr *mr)
3223 {
3224 	struct hns_roce_v2_mpt_entry *mpt_entry;
3225 
3226 	mpt_entry = mb_buf;
3227 	memset(mpt_entry, 0, sizeof(*mpt_entry));
3228 
3229 	hr_reg_write(mpt_entry, MPT_ST, V2_MPT_ST_VALID);
3230 	hr_reg_write(mpt_entry, MPT_PD, mr->pd);
3231 
3232 	hr_reg_write_bool(mpt_entry, MPT_BIND_EN,
3233 			  mr->access & IB_ACCESS_MW_BIND);
3234 	hr_reg_write_bool(mpt_entry, MPT_ATOMIC_EN,
3235 			  mr->access & IB_ACCESS_REMOTE_ATOMIC);
3236 	hr_reg_write_bool(mpt_entry, MPT_RR_EN,
3237 			  mr->access & IB_ACCESS_REMOTE_READ);
3238 	hr_reg_write_bool(mpt_entry, MPT_RW_EN,
3239 			  mr->access & IB_ACCESS_REMOTE_WRITE);
3240 	hr_reg_write_bool(mpt_entry, MPT_LW_EN,
3241 			  mr->access & IB_ACCESS_LOCAL_WRITE);
3242 
3243 	mpt_entry->len_l = cpu_to_le32(lower_32_bits(mr->size));
3244 	mpt_entry->len_h = cpu_to_le32(upper_32_bits(mr->size));
3245 	mpt_entry->lkey = cpu_to_le32(mr->key);
3246 	mpt_entry->va_l = cpu_to_le32(lower_32_bits(mr->iova));
3247 	mpt_entry->va_h = cpu_to_le32(upper_32_bits(mr->iova));
3248 
3249 	if (mr->type != MR_TYPE_MR)
3250 		hr_reg_enable(mpt_entry, MPT_PA);
3251 
3252 	if (mr->type == MR_TYPE_DMA)
3253 		return 0;
3254 
3255 	if (mr->pbl_hop_num != HNS_ROCE_HOP_NUM_0)
3256 		hr_reg_write(mpt_entry, MPT_PBL_HOP_NUM, mr->pbl_hop_num);
3257 
3258 	hr_reg_write(mpt_entry, MPT_PBL_BA_PG_SZ,
3259 		     to_hr_hw_page_shift(mr->pbl_mtr.hem_cfg.ba_pg_shift));
3260 	hr_reg_enable(mpt_entry, MPT_INNER_PA_VLD);
3261 
3262 	return set_mtpt_pbl(hr_dev, mpt_entry, mr);
3263 }
3264 
3265 static int hns_roce_v2_rereg_write_mtpt(struct hns_roce_dev *hr_dev,
3266 					struct hns_roce_mr *mr, int flags,
3267 					void *mb_buf)
3268 {
3269 	struct hns_roce_v2_mpt_entry *mpt_entry = mb_buf;
3270 	u32 mr_access_flags = mr->access;
3271 	int ret = 0;
3272 
3273 	hr_reg_write(mpt_entry, MPT_ST, V2_MPT_ST_VALID);
3274 	hr_reg_write(mpt_entry, MPT_PD, mr->pd);
3275 
3276 	if (flags & IB_MR_REREG_ACCESS) {
3277 		hr_reg_write(mpt_entry, MPT_BIND_EN,
3278 			     (mr_access_flags & IB_ACCESS_MW_BIND ? 1 : 0));
3279 		hr_reg_write(mpt_entry, MPT_ATOMIC_EN,
3280 			     mr_access_flags & IB_ACCESS_REMOTE_ATOMIC ? 1 : 0);
3281 		hr_reg_write(mpt_entry, MPT_RR_EN,
3282 			     mr_access_flags & IB_ACCESS_REMOTE_READ ? 1 : 0);
3283 		hr_reg_write(mpt_entry, MPT_RW_EN,
3284 			     mr_access_flags & IB_ACCESS_REMOTE_WRITE ? 1 : 0);
3285 		hr_reg_write(mpt_entry, MPT_LW_EN,
3286 			     mr_access_flags & IB_ACCESS_LOCAL_WRITE ? 1 : 0);
3287 	}
3288 
3289 	if (flags & IB_MR_REREG_TRANS) {
3290 		mpt_entry->va_l = cpu_to_le32(lower_32_bits(mr->iova));
3291 		mpt_entry->va_h = cpu_to_le32(upper_32_bits(mr->iova));
3292 		mpt_entry->len_l = cpu_to_le32(lower_32_bits(mr->size));
3293 		mpt_entry->len_h = cpu_to_le32(upper_32_bits(mr->size));
3294 
3295 		ret = set_mtpt_pbl(hr_dev, mpt_entry, mr);
3296 	}
3297 
3298 	return ret;
3299 }
3300 
3301 static int hns_roce_v2_frmr_write_mtpt(struct hns_roce_dev *hr_dev,
3302 				       void *mb_buf, struct hns_roce_mr *mr)
3303 {
3304 	struct ib_device *ibdev = &hr_dev->ib_dev;
3305 	struct hns_roce_v2_mpt_entry *mpt_entry;
3306 	dma_addr_t pbl_ba = 0;
3307 
3308 	mpt_entry = mb_buf;
3309 	memset(mpt_entry, 0, sizeof(*mpt_entry));
3310 
3311 	if (hns_roce_mtr_find(hr_dev, &mr->pbl_mtr, 0, NULL, 0, &pbl_ba) < 0) {
3312 		ibdev_err(ibdev, "failed to find frmr mtr.\n");
3313 		return -ENOBUFS;
3314 	}
3315 
3316 	hr_reg_write(mpt_entry, MPT_ST, V2_MPT_ST_FREE);
3317 	hr_reg_write(mpt_entry, MPT_PD, mr->pd);
3318 
3319 	hr_reg_enable(mpt_entry, MPT_RA_EN);
3320 	hr_reg_enable(mpt_entry, MPT_R_INV_EN);
3321 
3322 	hr_reg_enable(mpt_entry, MPT_FRE);
3323 	hr_reg_clear(mpt_entry, MPT_MR_MW);
3324 	hr_reg_enable(mpt_entry, MPT_BPD);
3325 	hr_reg_clear(mpt_entry, MPT_PA);
3326 
3327 	hr_reg_write(mpt_entry, MPT_PBL_HOP_NUM, 1);
3328 	hr_reg_write(mpt_entry, MPT_PBL_BA_PG_SZ,
3329 		     to_hr_hw_page_shift(mr->pbl_mtr.hem_cfg.ba_pg_shift));
3330 	hr_reg_write(mpt_entry, MPT_PBL_BUF_PG_SZ,
3331 		     to_hr_hw_page_shift(mr->pbl_mtr.hem_cfg.buf_pg_shift));
3332 
3333 	mpt_entry->pbl_size = cpu_to_le32(mr->npages);
3334 
3335 	mpt_entry->pbl_ba_l = cpu_to_le32(lower_32_bits(pbl_ba >> 3));
3336 	hr_reg_write(mpt_entry, MPT_PBL_BA_H, upper_32_bits(pbl_ba >> 3));
3337 
3338 	return 0;
3339 }
3340 
3341 static int hns_roce_v2_mw_write_mtpt(void *mb_buf, struct hns_roce_mw *mw)
3342 {
3343 	struct hns_roce_v2_mpt_entry *mpt_entry;
3344 
3345 	mpt_entry = mb_buf;
3346 	memset(mpt_entry, 0, sizeof(*mpt_entry));
3347 
3348 	hr_reg_write(mpt_entry, MPT_ST, V2_MPT_ST_FREE);
3349 	hr_reg_write(mpt_entry, MPT_PD, mw->pdn);
3350 
3351 	hr_reg_enable(mpt_entry, MPT_R_INV_EN);
3352 	hr_reg_enable(mpt_entry, MPT_LW_EN);
3353 
3354 	hr_reg_enable(mpt_entry, MPT_MR_MW);
3355 	hr_reg_enable(mpt_entry, MPT_BPD);
3356 	hr_reg_clear(mpt_entry, MPT_PA);
3357 	hr_reg_write(mpt_entry, MPT_BQP,
3358 		     mw->ibmw.type == IB_MW_TYPE_1 ? 0 : 1);
3359 
3360 	mpt_entry->lkey = cpu_to_le32(mw->rkey);
3361 
3362 	hr_reg_write(mpt_entry, MPT_PBL_HOP_NUM,
3363 		     mw->pbl_hop_num == HNS_ROCE_HOP_NUM_0 ? 0 :
3364 							     mw->pbl_hop_num);
3365 	hr_reg_write(mpt_entry, MPT_PBL_BA_PG_SZ,
3366 		     mw->pbl_ba_pg_sz + PG_SHIFT_OFFSET);
3367 	hr_reg_write(mpt_entry, MPT_PBL_BUF_PG_SZ,
3368 		     mw->pbl_buf_pg_sz + PG_SHIFT_OFFSET);
3369 
3370 	return 0;
3371 }
3372 
3373 static int free_mr_post_send_lp_wqe(struct hns_roce_qp *hr_qp)
3374 {
3375 	struct hns_roce_dev *hr_dev = to_hr_dev(hr_qp->ibqp.device);
3376 	struct ib_device *ibdev = &hr_dev->ib_dev;
3377 	const struct ib_send_wr *bad_wr;
3378 	struct ib_rdma_wr rdma_wr = {};
3379 	struct ib_send_wr *send_wr;
3380 	int ret;
3381 
3382 	send_wr = &rdma_wr.wr;
3383 	send_wr->opcode = IB_WR_RDMA_WRITE;
3384 
3385 	ret = hns_roce_v2_post_send(&hr_qp->ibqp, send_wr, &bad_wr);
3386 	if (ret) {
3387 		ibdev_err(ibdev, "failed to post wqe for free mr, ret = %d.\n",
3388 			  ret);
3389 		return ret;
3390 	}
3391 
3392 	return 0;
3393 }
3394 
3395 static int hns_roce_v2_poll_cq(struct ib_cq *ibcq, int num_entries,
3396 			       struct ib_wc *wc);
3397 
3398 static void free_mr_send_cmd_to_hw(struct hns_roce_dev *hr_dev)
3399 {
3400 	struct hns_roce_v2_priv *priv = hr_dev->priv;
3401 	struct hns_roce_v2_free_mr *free_mr = &priv->free_mr;
3402 	struct ib_wc wc[ARRAY_SIZE(free_mr->rsv_qp)];
3403 	struct ib_device *ibdev = &hr_dev->ib_dev;
3404 	struct hns_roce_qp *hr_qp;
3405 	unsigned long end;
3406 	int cqe_cnt = 0;
3407 	int npolled;
3408 	int ret;
3409 	int i;
3410 
3411 	/*
3412 	 * If the device initialization is not complete or in the uninstall
3413 	 * process, then there is no need to execute free mr.
3414 	 */
3415 	if (priv->handle->rinfo.reset_state == HNS_ROCE_STATE_RST_INIT ||
3416 	    priv->handle->rinfo.instance_state == HNS_ROCE_STATE_INIT ||
3417 	    hr_dev->state == HNS_ROCE_DEVICE_STATE_UNINIT)
3418 		return;
3419 
3420 	mutex_lock(&free_mr->mutex);
3421 
3422 	for (i = 0; i < ARRAY_SIZE(free_mr->rsv_qp); i++) {
3423 		hr_qp = free_mr->rsv_qp[i];
3424 
3425 		ret = free_mr_post_send_lp_wqe(hr_qp);
3426 		if (ret) {
3427 			ibdev_err(ibdev,
3428 				  "failed to send wqe (qp:0x%lx) for free mr, ret = %d.\n",
3429 				  hr_qp->qpn, ret);
3430 			break;
3431 		}
3432 
3433 		cqe_cnt++;
3434 	}
3435 
3436 	end = msecs_to_jiffies(HNS_ROCE_V2_FREE_MR_TIMEOUT) + jiffies;
3437 	while (cqe_cnt) {
3438 		npolled = hns_roce_v2_poll_cq(&free_mr->rsv_cq->ib_cq, cqe_cnt, wc);
3439 		if (npolled < 0) {
3440 			ibdev_err(ibdev,
3441 				  "failed to poll cqe for free mr, remain %d cqe.\n",
3442 				  cqe_cnt);
3443 			goto out;
3444 		}
3445 
3446 		if (time_after(jiffies, end)) {
3447 			ibdev_err(ibdev,
3448 				  "failed to poll cqe for free mr and timeout, remain %d cqe.\n",
3449 				  cqe_cnt);
3450 			goto out;
3451 		}
3452 		cqe_cnt -= npolled;
3453 	}
3454 
3455 out:
3456 	mutex_unlock(&free_mr->mutex);
3457 }
3458 
3459 static void hns_roce_v2_dereg_mr(struct hns_roce_dev *hr_dev)
3460 {
3461 	if (hr_dev->pci_dev->revision == PCI_REVISION_ID_HIP08)
3462 		free_mr_send_cmd_to_hw(hr_dev);
3463 }
3464 
3465 static void *get_cqe_v2(struct hns_roce_cq *hr_cq, int n)
3466 {
3467 	return hns_roce_buf_offset(hr_cq->mtr.kmem, n * hr_cq->cqe_size);
3468 }
3469 
3470 static void *get_sw_cqe_v2(struct hns_roce_cq *hr_cq, unsigned int n)
3471 {
3472 	struct hns_roce_v2_cqe *cqe = get_cqe_v2(hr_cq, n & hr_cq->ib_cq.cqe);
3473 
3474 	/* Get cqe when Owner bit is Conversely with the MSB of cons_idx */
3475 	return (hr_reg_read(cqe, CQE_OWNER) ^ !!(n & hr_cq->cq_depth)) ? cqe :
3476 									 NULL;
3477 }
3478 
3479 static inline void update_cq_db(struct hns_roce_dev *hr_dev,
3480 				struct hns_roce_cq *hr_cq)
3481 {
3482 	if (likely(hr_cq->flags & HNS_ROCE_CQ_FLAG_RECORD_DB)) {
3483 		*hr_cq->set_ci_db = hr_cq->cons_index & V2_CQ_DB_CONS_IDX_M;
3484 	} else {
3485 		struct hns_roce_v2_db cq_db = {};
3486 
3487 		hr_reg_write(&cq_db, DB_TAG, hr_cq->cqn);
3488 		hr_reg_write(&cq_db, DB_CMD, HNS_ROCE_V2_CQ_DB);
3489 		hr_reg_write(&cq_db, DB_CQ_CI, hr_cq->cons_index);
3490 		hr_reg_write(&cq_db, DB_CQ_CMD_SN, 1);
3491 
3492 		hns_roce_write64(hr_dev, (__le32 *)&cq_db, hr_cq->db_reg);
3493 	}
3494 }
3495 
3496 static void __hns_roce_v2_cq_clean(struct hns_roce_cq *hr_cq, u32 qpn,
3497 				   struct hns_roce_srq *srq)
3498 {
3499 	struct hns_roce_dev *hr_dev = to_hr_dev(hr_cq->ib_cq.device);
3500 	struct hns_roce_v2_cqe *cqe, *dest;
3501 	u32 prod_index;
3502 	int nfreed = 0;
3503 	int wqe_index;
3504 	u8 owner_bit;
3505 
3506 	for (prod_index = hr_cq->cons_index; get_sw_cqe_v2(hr_cq, prod_index);
3507 	     ++prod_index) {
3508 		if (prod_index > hr_cq->cons_index + hr_cq->ib_cq.cqe)
3509 			break;
3510 	}
3511 
3512 	/*
3513 	 * Now backwards through the CQ, removing CQ entries
3514 	 * that match our QP by overwriting them with next entries.
3515 	 */
3516 	while ((int) --prod_index - (int) hr_cq->cons_index >= 0) {
3517 		cqe = get_cqe_v2(hr_cq, prod_index & hr_cq->ib_cq.cqe);
3518 		if (hr_reg_read(cqe, CQE_LCL_QPN) == qpn) {
3519 			if (srq && hr_reg_read(cqe, CQE_S_R)) {
3520 				wqe_index = hr_reg_read(cqe, CQE_WQE_IDX);
3521 				hns_roce_free_srq_wqe(srq, wqe_index);
3522 			}
3523 			++nfreed;
3524 		} else if (nfreed) {
3525 			dest = get_cqe_v2(hr_cq, (prod_index + nfreed) &
3526 					  hr_cq->ib_cq.cqe);
3527 			owner_bit = hr_reg_read(dest, CQE_OWNER);
3528 			memcpy(dest, cqe, hr_cq->cqe_size);
3529 			hr_reg_write(dest, CQE_OWNER, owner_bit);
3530 		}
3531 	}
3532 
3533 	if (nfreed) {
3534 		hr_cq->cons_index += nfreed;
3535 		update_cq_db(hr_dev, hr_cq);
3536 	}
3537 }
3538 
3539 static void hns_roce_v2_cq_clean(struct hns_roce_cq *hr_cq, u32 qpn,
3540 				 struct hns_roce_srq *srq)
3541 {
3542 	spin_lock_irq(&hr_cq->lock);
3543 	__hns_roce_v2_cq_clean(hr_cq, qpn, srq);
3544 	spin_unlock_irq(&hr_cq->lock);
3545 }
3546 
3547 static void hns_roce_v2_write_cqc(struct hns_roce_dev *hr_dev,
3548 				  struct hns_roce_cq *hr_cq, void *mb_buf,
3549 				  u64 *mtts, dma_addr_t dma_handle)
3550 {
3551 	struct hns_roce_v2_cq_context *cq_context;
3552 
3553 	cq_context = mb_buf;
3554 	memset(cq_context, 0, sizeof(*cq_context));
3555 
3556 	hr_reg_write(cq_context, CQC_CQ_ST, V2_CQ_STATE_VALID);
3557 	hr_reg_write(cq_context, CQC_ARM_ST, NO_ARMED);
3558 	hr_reg_write(cq_context, CQC_SHIFT, ilog2(hr_cq->cq_depth));
3559 	hr_reg_write(cq_context, CQC_CEQN, hr_cq->vector);
3560 	hr_reg_write(cq_context, CQC_CQN, hr_cq->cqn);
3561 
3562 	if (hr_cq->cqe_size == HNS_ROCE_V3_CQE_SIZE)
3563 		hr_reg_write(cq_context, CQC_CQE_SIZE, CQE_SIZE_64B);
3564 
3565 	if (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_STASH)
3566 		hr_reg_enable(cq_context, CQC_STASH);
3567 
3568 	hr_reg_write(cq_context, CQC_CQE_CUR_BLK_ADDR_L,
3569 		     to_hr_hw_page_addr(mtts[0]));
3570 	hr_reg_write(cq_context, CQC_CQE_CUR_BLK_ADDR_H,
3571 		     upper_32_bits(to_hr_hw_page_addr(mtts[0])));
3572 	hr_reg_write(cq_context, CQC_CQE_HOP_NUM, hr_dev->caps.cqe_hop_num ==
3573 		     HNS_ROCE_HOP_NUM_0 ? 0 : hr_dev->caps.cqe_hop_num);
3574 	hr_reg_write(cq_context, CQC_CQE_NEX_BLK_ADDR_L,
3575 		     to_hr_hw_page_addr(mtts[1]));
3576 	hr_reg_write(cq_context, CQC_CQE_NEX_BLK_ADDR_H,
3577 		     upper_32_bits(to_hr_hw_page_addr(mtts[1])));
3578 	hr_reg_write(cq_context, CQC_CQE_BAR_PG_SZ,
3579 		     to_hr_hw_page_shift(hr_cq->mtr.hem_cfg.ba_pg_shift));
3580 	hr_reg_write(cq_context, CQC_CQE_BUF_PG_SZ,
3581 		     to_hr_hw_page_shift(hr_cq->mtr.hem_cfg.buf_pg_shift));
3582 	hr_reg_write(cq_context, CQC_CQE_BA_L, dma_handle >> 3);
3583 	hr_reg_write(cq_context, CQC_CQE_BA_H, (dma_handle >> (32 + 3)));
3584 	hr_reg_write_bool(cq_context, CQC_DB_RECORD_EN,
3585 			  hr_cq->flags & HNS_ROCE_CQ_FLAG_RECORD_DB);
3586 	hr_reg_write(cq_context, CQC_CQE_DB_RECORD_ADDR_L,
3587 		     ((u32)hr_cq->db.dma) >> 1);
3588 	hr_reg_write(cq_context, CQC_CQE_DB_RECORD_ADDR_H,
3589 		     hr_cq->db.dma >> 32);
3590 	hr_reg_write(cq_context, CQC_CQ_MAX_CNT,
3591 		     HNS_ROCE_V2_CQ_DEFAULT_BURST_NUM);
3592 	hr_reg_write(cq_context, CQC_CQ_PERIOD,
3593 		     HNS_ROCE_V2_CQ_DEFAULT_INTERVAL);
3594 }
3595 
3596 static int hns_roce_v2_req_notify_cq(struct ib_cq *ibcq,
3597 				     enum ib_cq_notify_flags flags)
3598 {
3599 	struct hns_roce_dev *hr_dev = to_hr_dev(ibcq->device);
3600 	struct hns_roce_cq *hr_cq = to_hr_cq(ibcq);
3601 	struct hns_roce_v2_db cq_db = {};
3602 	u32 notify_flag;
3603 
3604 	/*
3605 	 * flags = 0, then notify_flag : next
3606 	 * flags = 1, then notify flag : solocited
3607 	 */
3608 	notify_flag = (flags & IB_CQ_SOLICITED_MASK) == IB_CQ_SOLICITED ?
3609 		      V2_CQ_DB_REQ_NOT : V2_CQ_DB_REQ_NOT_SOL;
3610 
3611 	hr_reg_write(&cq_db, DB_TAG, hr_cq->cqn);
3612 	hr_reg_write(&cq_db, DB_CMD, HNS_ROCE_V2_CQ_DB_NOTIFY);
3613 	hr_reg_write(&cq_db, DB_CQ_CI, hr_cq->cons_index);
3614 	hr_reg_write(&cq_db, DB_CQ_CMD_SN, hr_cq->arm_sn);
3615 	hr_reg_write(&cq_db, DB_CQ_NOTIFY, notify_flag);
3616 
3617 	hns_roce_write64(hr_dev, (__le32 *)&cq_db, hr_cq->db_reg);
3618 
3619 	return 0;
3620 }
3621 
3622 static int sw_comp(struct hns_roce_qp *hr_qp, struct hns_roce_wq *wq,
3623 		   int num_entries, struct ib_wc *wc)
3624 {
3625 	unsigned int left;
3626 	int npolled = 0;
3627 
3628 	left = wq->head - wq->tail;
3629 	if (left == 0)
3630 		return 0;
3631 
3632 	left = min_t(unsigned int, (unsigned int)num_entries, left);
3633 	while (npolled < left) {
3634 		wc->wr_id = wq->wrid[wq->tail & (wq->wqe_cnt - 1)];
3635 		wc->status = IB_WC_WR_FLUSH_ERR;
3636 		wc->vendor_err = 0;
3637 		wc->qp = &hr_qp->ibqp;
3638 
3639 		wq->tail++;
3640 		wc++;
3641 		npolled++;
3642 	}
3643 
3644 	return npolled;
3645 }
3646 
3647 static int hns_roce_v2_sw_poll_cq(struct hns_roce_cq *hr_cq, int num_entries,
3648 				  struct ib_wc *wc)
3649 {
3650 	struct hns_roce_qp *hr_qp;
3651 	int npolled = 0;
3652 
3653 	list_for_each_entry(hr_qp, &hr_cq->sq_list, sq_node) {
3654 		npolled += sw_comp(hr_qp, &hr_qp->sq,
3655 				   num_entries - npolled, wc + npolled);
3656 		if (npolled >= num_entries)
3657 			goto out;
3658 	}
3659 
3660 	list_for_each_entry(hr_qp, &hr_cq->rq_list, rq_node) {
3661 		npolled += sw_comp(hr_qp, &hr_qp->rq,
3662 				   num_entries - npolled, wc + npolled);
3663 		if (npolled >= num_entries)
3664 			goto out;
3665 	}
3666 
3667 out:
3668 	return npolled;
3669 }
3670 
3671 static void get_cqe_status(struct hns_roce_dev *hr_dev, struct hns_roce_qp *qp,
3672 			   struct hns_roce_cq *cq, struct hns_roce_v2_cqe *cqe,
3673 			   struct ib_wc *wc)
3674 {
3675 	static const struct {
3676 		u32 cqe_status;
3677 		enum ib_wc_status wc_status;
3678 	} map[] = {
3679 		{ HNS_ROCE_CQE_V2_SUCCESS, IB_WC_SUCCESS },
3680 		{ HNS_ROCE_CQE_V2_LOCAL_LENGTH_ERR, IB_WC_LOC_LEN_ERR },
3681 		{ HNS_ROCE_CQE_V2_LOCAL_QP_OP_ERR, IB_WC_LOC_QP_OP_ERR },
3682 		{ HNS_ROCE_CQE_V2_LOCAL_PROT_ERR, IB_WC_LOC_PROT_ERR },
3683 		{ HNS_ROCE_CQE_V2_WR_FLUSH_ERR, IB_WC_WR_FLUSH_ERR },
3684 		{ HNS_ROCE_CQE_V2_MW_BIND_ERR, IB_WC_MW_BIND_ERR },
3685 		{ HNS_ROCE_CQE_V2_BAD_RESP_ERR, IB_WC_BAD_RESP_ERR },
3686 		{ HNS_ROCE_CQE_V2_LOCAL_ACCESS_ERR, IB_WC_LOC_ACCESS_ERR },
3687 		{ HNS_ROCE_CQE_V2_REMOTE_INVAL_REQ_ERR, IB_WC_REM_INV_REQ_ERR },
3688 		{ HNS_ROCE_CQE_V2_REMOTE_ACCESS_ERR, IB_WC_REM_ACCESS_ERR },
3689 		{ HNS_ROCE_CQE_V2_REMOTE_OP_ERR, IB_WC_REM_OP_ERR },
3690 		{ HNS_ROCE_CQE_V2_TRANSPORT_RETRY_EXC_ERR,
3691 		  IB_WC_RETRY_EXC_ERR },
3692 		{ HNS_ROCE_CQE_V2_RNR_RETRY_EXC_ERR, IB_WC_RNR_RETRY_EXC_ERR },
3693 		{ HNS_ROCE_CQE_V2_REMOTE_ABORT_ERR, IB_WC_REM_ABORT_ERR },
3694 		{ HNS_ROCE_CQE_V2_GENERAL_ERR, IB_WC_GENERAL_ERR}
3695 	};
3696 
3697 	u32 cqe_status = hr_reg_read(cqe, CQE_STATUS);
3698 	int i;
3699 
3700 	wc->status = IB_WC_GENERAL_ERR;
3701 	for (i = 0; i < ARRAY_SIZE(map); i++)
3702 		if (cqe_status == map[i].cqe_status) {
3703 			wc->status = map[i].wc_status;
3704 			break;
3705 		}
3706 
3707 	if (likely(wc->status == IB_WC_SUCCESS ||
3708 		   wc->status == IB_WC_WR_FLUSH_ERR))
3709 		return;
3710 
3711 	ibdev_err_ratelimited(&hr_dev->ib_dev, "error cqe status 0x%x:\n",
3712 			      cqe_status);
3713 	print_hex_dump(KERN_DEBUG, "", DUMP_PREFIX_NONE, 16, 4, cqe,
3714 		       cq->cqe_size, false);
3715 	wc->vendor_err = hr_reg_read(cqe, CQE_SUB_STATUS);
3716 
3717 	/*
3718 	 * For hns ROCEE, GENERAL_ERR is an error type that is not defined in
3719 	 * the standard protocol, the driver must ignore it and needn't to set
3720 	 * the QP to an error state.
3721 	 */
3722 	if (cqe_status == HNS_ROCE_CQE_V2_GENERAL_ERR)
3723 		return;
3724 
3725 	flush_cqe(hr_dev, qp);
3726 }
3727 
3728 static int get_cur_qp(struct hns_roce_cq *hr_cq, struct hns_roce_v2_cqe *cqe,
3729 		      struct hns_roce_qp **cur_qp)
3730 {
3731 	struct hns_roce_dev *hr_dev = to_hr_dev(hr_cq->ib_cq.device);
3732 	struct hns_roce_qp *hr_qp = *cur_qp;
3733 	u32 qpn;
3734 
3735 	qpn = hr_reg_read(cqe, CQE_LCL_QPN);
3736 
3737 	if (!hr_qp || qpn != hr_qp->qpn) {
3738 		hr_qp = __hns_roce_qp_lookup(hr_dev, qpn);
3739 		if (unlikely(!hr_qp)) {
3740 			ibdev_err(&hr_dev->ib_dev,
3741 				  "CQ %06lx with entry for unknown QPN %06x\n",
3742 				  hr_cq->cqn, qpn);
3743 			return -EINVAL;
3744 		}
3745 		*cur_qp = hr_qp;
3746 	}
3747 
3748 	return 0;
3749 }
3750 
3751 /*
3752  * mapped-value = 1 + real-value
3753  * The ib wc opcode's real value is start from 0, In order to distinguish
3754  * between initialized and uninitialized map values, we plus 1 to the actual
3755  * value when defining the mapping, so that the validity can be identified by
3756  * checking whether the mapped value is greater than 0.
3757  */
3758 #define HR_WC_OP_MAP(hr_key, ib_key) \
3759 		[HNS_ROCE_V2_WQE_OP_ ## hr_key] = 1 + IB_WC_ ## ib_key
3760 
3761 static const u32 wc_send_op_map[] = {
3762 	HR_WC_OP_MAP(SEND,			SEND),
3763 	HR_WC_OP_MAP(SEND_WITH_INV,		SEND),
3764 	HR_WC_OP_MAP(SEND_WITH_IMM,		SEND),
3765 	HR_WC_OP_MAP(RDMA_READ,			RDMA_READ),
3766 	HR_WC_OP_MAP(RDMA_WRITE,		RDMA_WRITE),
3767 	HR_WC_OP_MAP(RDMA_WRITE_WITH_IMM,	RDMA_WRITE),
3768 	HR_WC_OP_MAP(ATOM_CMP_AND_SWAP,		COMP_SWAP),
3769 	HR_WC_OP_MAP(ATOM_FETCH_AND_ADD,	FETCH_ADD),
3770 	HR_WC_OP_MAP(ATOM_MSK_CMP_AND_SWAP,	MASKED_COMP_SWAP),
3771 	HR_WC_OP_MAP(ATOM_MSK_FETCH_AND_ADD,	MASKED_FETCH_ADD),
3772 	HR_WC_OP_MAP(FAST_REG_PMR,		REG_MR),
3773 	HR_WC_OP_MAP(BIND_MW,			REG_MR),
3774 };
3775 
3776 static int to_ib_wc_send_op(u32 hr_opcode)
3777 {
3778 	if (hr_opcode >= ARRAY_SIZE(wc_send_op_map))
3779 		return -EINVAL;
3780 
3781 	return wc_send_op_map[hr_opcode] ? wc_send_op_map[hr_opcode] - 1 :
3782 					   -EINVAL;
3783 }
3784 
3785 static const u32 wc_recv_op_map[] = {
3786 	HR_WC_OP_MAP(RDMA_WRITE_WITH_IMM,		WITH_IMM),
3787 	HR_WC_OP_MAP(SEND,				RECV),
3788 	HR_WC_OP_MAP(SEND_WITH_IMM,			WITH_IMM),
3789 	HR_WC_OP_MAP(SEND_WITH_INV,			RECV),
3790 };
3791 
3792 static int to_ib_wc_recv_op(u32 hr_opcode)
3793 {
3794 	if (hr_opcode >= ARRAY_SIZE(wc_recv_op_map))
3795 		return -EINVAL;
3796 
3797 	return wc_recv_op_map[hr_opcode] ? wc_recv_op_map[hr_opcode] - 1 :
3798 					   -EINVAL;
3799 }
3800 
3801 static void fill_send_wc(struct ib_wc *wc, struct hns_roce_v2_cqe *cqe)
3802 {
3803 	u32 hr_opcode;
3804 	int ib_opcode;
3805 
3806 	wc->wc_flags = 0;
3807 
3808 	hr_opcode = hr_reg_read(cqe, CQE_OPCODE);
3809 	switch (hr_opcode) {
3810 	case HNS_ROCE_V2_WQE_OP_RDMA_READ:
3811 		wc->byte_len = le32_to_cpu(cqe->byte_cnt);
3812 		break;
3813 	case HNS_ROCE_V2_WQE_OP_SEND_WITH_IMM:
3814 	case HNS_ROCE_V2_WQE_OP_RDMA_WRITE_WITH_IMM:
3815 		wc->wc_flags |= IB_WC_WITH_IMM;
3816 		break;
3817 	case HNS_ROCE_V2_WQE_OP_ATOM_CMP_AND_SWAP:
3818 	case HNS_ROCE_V2_WQE_OP_ATOM_FETCH_AND_ADD:
3819 	case HNS_ROCE_V2_WQE_OP_ATOM_MSK_CMP_AND_SWAP:
3820 	case HNS_ROCE_V2_WQE_OP_ATOM_MSK_FETCH_AND_ADD:
3821 		wc->byte_len  = 8;
3822 		break;
3823 	default:
3824 		break;
3825 	}
3826 
3827 	ib_opcode = to_ib_wc_send_op(hr_opcode);
3828 	if (ib_opcode < 0)
3829 		wc->status = IB_WC_GENERAL_ERR;
3830 	else
3831 		wc->opcode = ib_opcode;
3832 }
3833 
3834 static int fill_recv_wc(struct ib_wc *wc, struct hns_roce_v2_cqe *cqe)
3835 {
3836 	u32 hr_opcode;
3837 	int ib_opcode;
3838 
3839 	wc->byte_len = le32_to_cpu(cqe->byte_cnt);
3840 
3841 	hr_opcode = hr_reg_read(cqe, CQE_OPCODE);
3842 	switch (hr_opcode) {
3843 	case HNS_ROCE_V2_OPCODE_RDMA_WRITE_IMM:
3844 	case HNS_ROCE_V2_OPCODE_SEND_WITH_IMM:
3845 		wc->wc_flags = IB_WC_WITH_IMM;
3846 		wc->ex.imm_data = cpu_to_be32(le32_to_cpu(cqe->immtdata));
3847 		break;
3848 	case HNS_ROCE_V2_OPCODE_SEND_WITH_INV:
3849 		wc->wc_flags = IB_WC_WITH_INVALIDATE;
3850 		wc->ex.invalidate_rkey = le32_to_cpu(cqe->rkey);
3851 		break;
3852 	default:
3853 		wc->wc_flags = 0;
3854 	}
3855 
3856 	ib_opcode = to_ib_wc_recv_op(hr_opcode);
3857 	if (ib_opcode < 0)
3858 		wc->status = IB_WC_GENERAL_ERR;
3859 	else
3860 		wc->opcode = ib_opcode;
3861 
3862 	wc->sl = hr_reg_read(cqe, CQE_SL);
3863 	wc->src_qp = hr_reg_read(cqe, CQE_RMT_QPN);
3864 	wc->slid = 0;
3865 	wc->wc_flags |= hr_reg_read(cqe, CQE_GRH) ? IB_WC_GRH : 0;
3866 	wc->port_num = hr_reg_read(cqe, CQE_PORTN);
3867 	wc->pkey_index = 0;
3868 
3869 	if (hr_reg_read(cqe, CQE_VID_VLD)) {
3870 		wc->vlan_id = hr_reg_read(cqe, CQE_VID);
3871 		wc->wc_flags |= IB_WC_WITH_VLAN;
3872 	} else {
3873 		wc->vlan_id = 0xffff;
3874 	}
3875 
3876 	wc->network_hdr_type = hr_reg_read(cqe, CQE_PORT_TYPE);
3877 
3878 	return 0;
3879 }
3880 
3881 static int hns_roce_v2_poll_one(struct hns_roce_cq *hr_cq,
3882 				struct hns_roce_qp **cur_qp, struct ib_wc *wc)
3883 {
3884 	struct hns_roce_dev *hr_dev = to_hr_dev(hr_cq->ib_cq.device);
3885 	struct hns_roce_qp *qp = *cur_qp;
3886 	struct hns_roce_srq *srq = NULL;
3887 	struct hns_roce_v2_cqe *cqe;
3888 	struct hns_roce_wq *wq;
3889 	int is_send;
3890 	u16 wqe_idx;
3891 	int ret;
3892 
3893 	cqe = get_sw_cqe_v2(hr_cq, hr_cq->cons_index);
3894 	if (!cqe)
3895 		return -EAGAIN;
3896 
3897 	++hr_cq->cons_index;
3898 	/* Memory barrier */
3899 	rmb();
3900 
3901 	ret = get_cur_qp(hr_cq, cqe, &qp);
3902 	if (ret)
3903 		return ret;
3904 
3905 	wc->qp = &qp->ibqp;
3906 	wc->vendor_err = 0;
3907 
3908 	wqe_idx = hr_reg_read(cqe, CQE_WQE_IDX);
3909 
3910 	is_send = !hr_reg_read(cqe, CQE_S_R);
3911 	if (is_send) {
3912 		wq = &qp->sq;
3913 
3914 		/* If sg_signal_bit is set, tail pointer will be updated to
3915 		 * the WQE corresponding to the current CQE.
3916 		 */
3917 		if (qp->sq_signal_bits)
3918 			wq->tail += (wqe_idx - (u16)wq->tail) &
3919 				    (wq->wqe_cnt - 1);
3920 
3921 		wc->wr_id = wq->wrid[wq->tail & (wq->wqe_cnt - 1)];
3922 		++wq->tail;
3923 
3924 		fill_send_wc(wc, cqe);
3925 	} else {
3926 		if (qp->ibqp.srq) {
3927 			srq = to_hr_srq(qp->ibqp.srq);
3928 			wc->wr_id = srq->wrid[wqe_idx];
3929 			hns_roce_free_srq_wqe(srq, wqe_idx);
3930 		} else {
3931 			wq = &qp->rq;
3932 			wc->wr_id = wq->wrid[wq->tail & (wq->wqe_cnt - 1)];
3933 			++wq->tail;
3934 		}
3935 
3936 		ret = fill_recv_wc(wc, cqe);
3937 	}
3938 
3939 	get_cqe_status(hr_dev, qp, hr_cq, cqe, wc);
3940 	if (unlikely(wc->status != IB_WC_SUCCESS))
3941 		return 0;
3942 
3943 	return ret;
3944 }
3945 
3946 static int hns_roce_v2_poll_cq(struct ib_cq *ibcq, int num_entries,
3947 			       struct ib_wc *wc)
3948 {
3949 	struct hns_roce_dev *hr_dev = to_hr_dev(ibcq->device);
3950 	struct hns_roce_cq *hr_cq = to_hr_cq(ibcq);
3951 	struct hns_roce_qp *cur_qp = NULL;
3952 	unsigned long flags;
3953 	int npolled;
3954 
3955 	spin_lock_irqsave(&hr_cq->lock, flags);
3956 
3957 	/*
3958 	 * When the device starts to reset, the state is RST_DOWN. At this time,
3959 	 * there may still be some valid CQEs in the hardware that are not
3960 	 * polled. Therefore, it is not allowed to switch to the software mode
3961 	 * immediately. When the state changes to UNINIT, CQE no longer exists
3962 	 * in the hardware, and then switch to software mode.
3963 	 */
3964 	if (hr_dev->state == HNS_ROCE_DEVICE_STATE_UNINIT) {
3965 		npolled = hns_roce_v2_sw_poll_cq(hr_cq, num_entries, wc);
3966 		goto out;
3967 	}
3968 
3969 	for (npolled = 0; npolled < num_entries; ++npolled) {
3970 		if (hns_roce_v2_poll_one(hr_cq, &cur_qp, wc + npolled))
3971 			break;
3972 	}
3973 
3974 	if (npolled)
3975 		update_cq_db(hr_dev, hr_cq);
3976 
3977 out:
3978 	spin_unlock_irqrestore(&hr_cq->lock, flags);
3979 
3980 	return npolled;
3981 }
3982 
3983 static int get_op_for_set_hem(struct hns_roce_dev *hr_dev, u32 type,
3984 			      u32 step_idx, u8 *mbox_cmd)
3985 {
3986 	u8 cmd;
3987 
3988 	switch (type) {
3989 	case HEM_TYPE_QPC:
3990 		cmd = HNS_ROCE_CMD_WRITE_QPC_BT0;
3991 		break;
3992 	case HEM_TYPE_MTPT:
3993 		cmd = HNS_ROCE_CMD_WRITE_MPT_BT0;
3994 		break;
3995 	case HEM_TYPE_CQC:
3996 		cmd = HNS_ROCE_CMD_WRITE_CQC_BT0;
3997 		break;
3998 	case HEM_TYPE_SRQC:
3999 		cmd = HNS_ROCE_CMD_WRITE_SRQC_BT0;
4000 		break;
4001 	case HEM_TYPE_SCCC:
4002 		cmd = HNS_ROCE_CMD_WRITE_SCCC_BT0;
4003 		break;
4004 	case HEM_TYPE_QPC_TIMER:
4005 		cmd = HNS_ROCE_CMD_WRITE_QPC_TIMER_BT0;
4006 		break;
4007 	case HEM_TYPE_CQC_TIMER:
4008 		cmd = HNS_ROCE_CMD_WRITE_CQC_TIMER_BT0;
4009 		break;
4010 	default:
4011 		dev_warn(hr_dev->dev, "failed to check hem type %u.\n", type);
4012 		return -EINVAL;
4013 	}
4014 
4015 	*mbox_cmd = cmd + step_idx;
4016 
4017 	return 0;
4018 }
4019 
4020 static int config_gmv_ba_to_hw(struct hns_roce_dev *hr_dev, unsigned long obj,
4021 			       dma_addr_t base_addr)
4022 {
4023 	struct hns_roce_cmq_desc desc;
4024 	struct hns_roce_cmq_req *req = (struct hns_roce_cmq_req *)desc.data;
4025 	u32 idx = obj / (HNS_HW_PAGE_SIZE / hr_dev->caps.gmv_entry_sz);
4026 	u64 addr = to_hr_hw_page_addr(base_addr);
4027 
4028 	hns_roce_cmq_setup_basic_desc(&desc, HNS_ROCE_OPC_CFG_GMV_BT, false);
4029 
4030 	hr_reg_write(req, CFG_GMV_BT_BA_L, lower_32_bits(addr));
4031 	hr_reg_write(req, CFG_GMV_BT_BA_H, upper_32_bits(addr));
4032 	hr_reg_write(req, CFG_GMV_BT_IDX, idx);
4033 
4034 	return hns_roce_cmq_send(hr_dev, &desc, 1);
4035 }
4036 
4037 static int set_hem_to_hw(struct hns_roce_dev *hr_dev, int obj,
4038 			 dma_addr_t base_addr, u32 hem_type, u32 step_idx)
4039 {
4040 	int ret;
4041 	u8 cmd;
4042 
4043 	if (unlikely(hem_type == HEM_TYPE_GMV))
4044 		return config_gmv_ba_to_hw(hr_dev, obj, base_addr);
4045 
4046 	if (unlikely(hem_type == HEM_TYPE_SCCC && step_idx))
4047 		return 0;
4048 
4049 	ret = get_op_for_set_hem(hr_dev, hem_type, step_idx, &cmd);
4050 	if (ret < 0)
4051 		return ret;
4052 
4053 	return config_hem_ba_to_hw(hr_dev, base_addr, cmd, obj);
4054 }
4055 
4056 static int hns_roce_v2_set_hem(struct hns_roce_dev *hr_dev,
4057 			       struct hns_roce_hem_table *table, int obj,
4058 			       u32 step_idx)
4059 {
4060 	struct hns_roce_hem_iter iter;
4061 	struct hns_roce_hem_mhop mhop;
4062 	struct hns_roce_hem *hem;
4063 	unsigned long mhop_obj = obj;
4064 	int i, j, k;
4065 	int ret = 0;
4066 	u64 hem_idx = 0;
4067 	u64 l1_idx = 0;
4068 	u64 bt_ba = 0;
4069 	u32 chunk_ba_num;
4070 	u32 hop_num;
4071 
4072 	if (!hns_roce_check_whether_mhop(hr_dev, table->type))
4073 		return 0;
4074 
4075 	hns_roce_calc_hem_mhop(hr_dev, table, &mhop_obj, &mhop);
4076 	i = mhop.l0_idx;
4077 	j = mhop.l1_idx;
4078 	k = mhop.l2_idx;
4079 	hop_num = mhop.hop_num;
4080 	chunk_ba_num = mhop.bt_chunk_size / 8;
4081 
4082 	if (hop_num == 2) {
4083 		hem_idx = i * chunk_ba_num * chunk_ba_num + j * chunk_ba_num +
4084 			  k;
4085 		l1_idx = i * chunk_ba_num + j;
4086 	} else if (hop_num == 1) {
4087 		hem_idx = i * chunk_ba_num + j;
4088 	} else if (hop_num == HNS_ROCE_HOP_NUM_0) {
4089 		hem_idx = i;
4090 	}
4091 
4092 	if (table->type == HEM_TYPE_SCCC)
4093 		obj = mhop.l0_idx;
4094 
4095 	if (check_whether_last_step(hop_num, step_idx)) {
4096 		hem = table->hem[hem_idx];
4097 		for (hns_roce_hem_first(hem, &iter);
4098 		     !hns_roce_hem_last(&iter); hns_roce_hem_next(&iter)) {
4099 			bt_ba = hns_roce_hem_addr(&iter);
4100 			ret = set_hem_to_hw(hr_dev, obj, bt_ba, table->type,
4101 					    step_idx);
4102 		}
4103 	} else {
4104 		if (step_idx == 0)
4105 			bt_ba = table->bt_l0_dma_addr[i];
4106 		else if (step_idx == 1 && hop_num == 2)
4107 			bt_ba = table->bt_l1_dma_addr[l1_idx];
4108 
4109 		ret = set_hem_to_hw(hr_dev, obj, bt_ba, table->type, step_idx);
4110 	}
4111 
4112 	return ret;
4113 }
4114 
4115 static int hns_roce_v2_clear_hem(struct hns_roce_dev *hr_dev,
4116 				 struct hns_roce_hem_table *table,
4117 				 int tag, u32 step_idx)
4118 {
4119 	struct hns_roce_cmd_mailbox *mailbox;
4120 	struct device *dev = hr_dev->dev;
4121 	u8 cmd = 0xff;
4122 	int ret;
4123 
4124 	if (!hns_roce_check_whether_mhop(hr_dev, table->type))
4125 		return 0;
4126 
4127 	switch (table->type) {
4128 	case HEM_TYPE_QPC:
4129 		cmd = HNS_ROCE_CMD_DESTROY_QPC_BT0;
4130 		break;
4131 	case HEM_TYPE_MTPT:
4132 		cmd = HNS_ROCE_CMD_DESTROY_MPT_BT0;
4133 		break;
4134 	case HEM_TYPE_CQC:
4135 		cmd = HNS_ROCE_CMD_DESTROY_CQC_BT0;
4136 		break;
4137 	case HEM_TYPE_SRQC:
4138 		cmd = HNS_ROCE_CMD_DESTROY_SRQC_BT0;
4139 		break;
4140 	case HEM_TYPE_SCCC:
4141 	case HEM_TYPE_QPC_TIMER:
4142 	case HEM_TYPE_CQC_TIMER:
4143 	case HEM_TYPE_GMV:
4144 		return 0;
4145 	default:
4146 		dev_warn(dev, "table %u not to be destroyed by mailbox!\n",
4147 			 table->type);
4148 		return 0;
4149 	}
4150 
4151 	cmd += step_idx;
4152 
4153 	mailbox = hns_roce_alloc_cmd_mailbox(hr_dev);
4154 	if (IS_ERR(mailbox))
4155 		return PTR_ERR(mailbox);
4156 
4157 	ret = hns_roce_cmd_mbox(hr_dev, 0, mailbox->dma, cmd, tag);
4158 
4159 	hns_roce_free_cmd_mailbox(hr_dev, mailbox);
4160 	return ret;
4161 }
4162 
4163 static int hns_roce_v2_qp_modify(struct hns_roce_dev *hr_dev,
4164 				 struct hns_roce_v2_qp_context *context,
4165 				 struct hns_roce_v2_qp_context *qpc_mask,
4166 				 struct hns_roce_qp *hr_qp)
4167 {
4168 	struct hns_roce_cmd_mailbox *mailbox;
4169 	int qpc_size;
4170 	int ret;
4171 
4172 	mailbox = hns_roce_alloc_cmd_mailbox(hr_dev);
4173 	if (IS_ERR(mailbox))
4174 		return PTR_ERR(mailbox);
4175 
4176 	/* The qpc size of HIP08 is only 256B, which is half of HIP09 */
4177 	qpc_size = hr_dev->caps.qpc_sz;
4178 	memcpy(mailbox->buf, context, qpc_size);
4179 	memcpy(mailbox->buf + qpc_size, qpc_mask, qpc_size);
4180 
4181 	ret = hns_roce_cmd_mbox(hr_dev, mailbox->dma, 0,
4182 				HNS_ROCE_CMD_MODIFY_QPC, hr_qp->qpn);
4183 
4184 	hns_roce_free_cmd_mailbox(hr_dev, mailbox);
4185 
4186 	return ret;
4187 }
4188 
4189 static void set_access_flags(struct hns_roce_qp *hr_qp,
4190 			     struct hns_roce_v2_qp_context *context,
4191 			     struct hns_roce_v2_qp_context *qpc_mask,
4192 			     const struct ib_qp_attr *attr, int attr_mask)
4193 {
4194 	u8 dest_rd_atomic;
4195 	u32 access_flags;
4196 
4197 	dest_rd_atomic = (attr_mask & IB_QP_MAX_DEST_RD_ATOMIC) ?
4198 			 attr->max_dest_rd_atomic : hr_qp->resp_depth;
4199 
4200 	access_flags = (attr_mask & IB_QP_ACCESS_FLAGS) ?
4201 		       attr->qp_access_flags : hr_qp->atomic_rd_en;
4202 
4203 	if (!dest_rd_atomic)
4204 		access_flags &= IB_ACCESS_REMOTE_WRITE;
4205 
4206 	hr_reg_write_bool(context, QPC_RRE,
4207 			  access_flags & IB_ACCESS_REMOTE_READ);
4208 	hr_reg_clear(qpc_mask, QPC_RRE);
4209 
4210 	hr_reg_write_bool(context, QPC_RWE,
4211 			  access_flags & IB_ACCESS_REMOTE_WRITE);
4212 	hr_reg_clear(qpc_mask, QPC_RWE);
4213 
4214 	hr_reg_write_bool(context, QPC_ATE,
4215 			  access_flags & IB_ACCESS_REMOTE_ATOMIC);
4216 	hr_reg_clear(qpc_mask, QPC_ATE);
4217 	hr_reg_write_bool(context, QPC_EXT_ATE,
4218 			  access_flags & IB_ACCESS_REMOTE_ATOMIC);
4219 	hr_reg_clear(qpc_mask, QPC_EXT_ATE);
4220 }
4221 
4222 static void set_qpc_wqe_cnt(struct hns_roce_qp *hr_qp,
4223 			    struct hns_roce_v2_qp_context *context,
4224 			    struct hns_roce_v2_qp_context *qpc_mask)
4225 {
4226 	hr_reg_write(context, QPC_SGE_SHIFT,
4227 		     to_hr_hem_entries_shift(hr_qp->sge.sge_cnt,
4228 					     hr_qp->sge.sge_shift));
4229 
4230 	hr_reg_write(context, QPC_SQ_SHIFT, ilog2(hr_qp->sq.wqe_cnt));
4231 
4232 	hr_reg_write(context, QPC_RQ_SHIFT, ilog2(hr_qp->rq.wqe_cnt));
4233 }
4234 
4235 static inline int get_cqn(struct ib_cq *ib_cq)
4236 {
4237 	return ib_cq ? to_hr_cq(ib_cq)->cqn : 0;
4238 }
4239 
4240 static inline int get_pdn(struct ib_pd *ib_pd)
4241 {
4242 	return ib_pd ? to_hr_pd(ib_pd)->pdn : 0;
4243 }
4244 
4245 static void modify_qp_reset_to_init(struct ib_qp *ibqp,
4246 				    const struct ib_qp_attr *attr,
4247 				    struct hns_roce_v2_qp_context *context,
4248 				    struct hns_roce_v2_qp_context *qpc_mask)
4249 {
4250 	struct hns_roce_dev *hr_dev = to_hr_dev(ibqp->device);
4251 	struct hns_roce_qp *hr_qp = to_hr_qp(ibqp);
4252 
4253 	/*
4254 	 * In v2 engine, software pass context and context mask to hardware
4255 	 * when modifying qp. If software need modify some fields in context,
4256 	 * we should set all bits of the relevant fields in context mask to
4257 	 * 0 at the same time, else set them to 0x1.
4258 	 */
4259 	hr_reg_write(context, QPC_TST, to_hr_qp_type(ibqp->qp_type));
4260 
4261 	hr_reg_write(context, QPC_PD, get_pdn(ibqp->pd));
4262 
4263 	hr_reg_write(context, QPC_RQWS, ilog2(hr_qp->rq.max_gs));
4264 
4265 	set_qpc_wqe_cnt(hr_qp, context, qpc_mask);
4266 
4267 	/* No VLAN need to set 0xFFF */
4268 	hr_reg_write(context, QPC_VLAN_ID, 0xfff);
4269 
4270 	if (ibqp->qp_type == IB_QPT_XRC_TGT) {
4271 		context->qkey_xrcd = cpu_to_le32(hr_qp->xrcdn);
4272 
4273 		hr_reg_enable(context, QPC_XRC_QP_TYPE);
4274 	}
4275 
4276 	if (hr_qp->en_flags & HNS_ROCE_QP_CAP_RQ_RECORD_DB)
4277 		hr_reg_enable(context, QPC_RQ_RECORD_EN);
4278 
4279 	if (hr_qp->en_flags & HNS_ROCE_QP_CAP_OWNER_DB)
4280 		hr_reg_enable(context, QPC_OWNER_MODE);
4281 
4282 	hr_reg_write(context, QPC_RQ_DB_RECORD_ADDR_L,
4283 		     lower_32_bits(hr_qp->rdb.dma) >> 1);
4284 	hr_reg_write(context, QPC_RQ_DB_RECORD_ADDR_H,
4285 		     upper_32_bits(hr_qp->rdb.dma));
4286 
4287 	hr_reg_write(context, QPC_RX_CQN, get_cqn(ibqp->recv_cq));
4288 
4289 	if (ibqp->srq) {
4290 		hr_reg_enable(context, QPC_SRQ_EN);
4291 		hr_reg_write(context, QPC_SRQN, to_hr_srq(ibqp->srq)->srqn);
4292 	}
4293 
4294 	hr_reg_enable(context, QPC_FRE);
4295 
4296 	hr_reg_write(context, QPC_TX_CQN, get_cqn(ibqp->send_cq));
4297 
4298 	if (hr_dev->caps.qpc_sz < HNS_ROCE_V3_QPC_SZ)
4299 		return;
4300 
4301 	if (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_STASH)
4302 		hr_reg_enable(&context->ext, QPCEX_STASH);
4303 }
4304 
4305 static void modify_qp_init_to_init(struct ib_qp *ibqp,
4306 				   const struct ib_qp_attr *attr,
4307 				   struct hns_roce_v2_qp_context *context,
4308 				   struct hns_roce_v2_qp_context *qpc_mask)
4309 {
4310 	/*
4311 	 * In v2 engine, software pass context and context mask to hardware
4312 	 * when modifying qp. If software need modify some fields in context,
4313 	 * we should set all bits of the relevant fields in context mask to
4314 	 * 0 at the same time, else set them to 0x1.
4315 	 */
4316 	hr_reg_write(context, QPC_TST, to_hr_qp_type(ibqp->qp_type));
4317 	hr_reg_clear(qpc_mask, QPC_TST);
4318 
4319 	hr_reg_write(context, QPC_PD, get_pdn(ibqp->pd));
4320 	hr_reg_clear(qpc_mask, QPC_PD);
4321 
4322 	hr_reg_write(context, QPC_RX_CQN, get_cqn(ibqp->recv_cq));
4323 	hr_reg_clear(qpc_mask, QPC_RX_CQN);
4324 
4325 	hr_reg_write(context, QPC_TX_CQN, get_cqn(ibqp->send_cq));
4326 	hr_reg_clear(qpc_mask, QPC_TX_CQN);
4327 
4328 	if (ibqp->srq) {
4329 		hr_reg_enable(context, QPC_SRQ_EN);
4330 		hr_reg_clear(qpc_mask, QPC_SRQ_EN);
4331 		hr_reg_write(context, QPC_SRQN, to_hr_srq(ibqp->srq)->srqn);
4332 		hr_reg_clear(qpc_mask, QPC_SRQN);
4333 	}
4334 }
4335 
4336 static int config_qp_rq_buf(struct hns_roce_dev *hr_dev,
4337 			    struct hns_roce_qp *hr_qp,
4338 			    struct hns_roce_v2_qp_context *context,
4339 			    struct hns_roce_v2_qp_context *qpc_mask)
4340 {
4341 	u64 mtts[MTT_MIN_COUNT] = { 0 };
4342 	u64 wqe_sge_ba;
4343 	int count;
4344 
4345 	/* Search qp buf's mtts */
4346 	count = hns_roce_mtr_find(hr_dev, &hr_qp->mtr, hr_qp->rq.offset, mtts,
4347 				  MTT_MIN_COUNT, &wqe_sge_ba);
4348 	if (hr_qp->rq.wqe_cnt && count < 1) {
4349 		ibdev_err(&hr_dev->ib_dev,
4350 			  "failed to find RQ WQE, QPN = 0x%lx.\n", hr_qp->qpn);
4351 		return -EINVAL;
4352 	}
4353 
4354 	context->wqe_sge_ba = cpu_to_le32(wqe_sge_ba >> 3);
4355 	qpc_mask->wqe_sge_ba = 0;
4356 
4357 	/*
4358 	 * In v2 engine, software pass context and context mask to hardware
4359 	 * when modifying qp. If software need modify some fields in context,
4360 	 * we should set all bits of the relevant fields in context mask to
4361 	 * 0 at the same time, else set them to 0x1.
4362 	 */
4363 	hr_reg_write(context, QPC_WQE_SGE_BA_H, wqe_sge_ba >> (32 + 3));
4364 	hr_reg_clear(qpc_mask, QPC_WQE_SGE_BA_H);
4365 
4366 	hr_reg_write(context, QPC_SQ_HOP_NUM,
4367 		     to_hr_hem_hopnum(hr_dev->caps.wqe_sq_hop_num,
4368 				      hr_qp->sq.wqe_cnt));
4369 	hr_reg_clear(qpc_mask, QPC_SQ_HOP_NUM);
4370 
4371 	hr_reg_write(context, QPC_SGE_HOP_NUM,
4372 		     to_hr_hem_hopnum(hr_dev->caps.wqe_sge_hop_num,
4373 				      hr_qp->sge.sge_cnt));
4374 	hr_reg_clear(qpc_mask, QPC_SGE_HOP_NUM);
4375 
4376 	hr_reg_write(context, QPC_RQ_HOP_NUM,
4377 		     to_hr_hem_hopnum(hr_dev->caps.wqe_rq_hop_num,
4378 				      hr_qp->rq.wqe_cnt));
4379 
4380 	hr_reg_clear(qpc_mask, QPC_RQ_HOP_NUM);
4381 
4382 	hr_reg_write(context, QPC_WQE_SGE_BA_PG_SZ,
4383 		     to_hr_hw_page_shift(hr_qp->mtr.hem_cfg.ba_pg_shift));
4384 	hr_reg_clear(qpc_mask, QPC_WQE_SGE_BA_PG_SZ);
4385 
4386 	hr_reg_write(context, QPC_WQE_SGE_BUF_PG_SZ,
4387 		     to_hr_hw_page_shift(hr_qp->mtr.hem_cfg.buf_pg_shift));
4388 	hr_reg_clear(qpc_mask, QPC_WQE_SGE_BUF_PG_SZ);
4389 
4390 	context->rq_cur_blk_addr = cpu_to_le32(to_hr_hw_page_addr(mtts[0]));
4391 	qpc_mask->rq_cur_blk_addr = 0;
4392 
4393 	hr_reg_write(context, QPC_RQ_CUR_BLK_ADDR_H,
4394 		     upper_32_bits(to_hr_hw_page_addr(mtts[0])));
4395 	hr_reg_clear(qpc_mask, QPC_RQ_CUR_BLK_ADDR_H);
4396 
4397 	if (hr_dev->pci_dev->revision == PCI_REVISION_ID_HIP08) {
4398 		context->rq_nxt_blk_addr =
4399 				cpu_to_le32(to_hr_hw_page_addr(mtts[1]));
4400 		qpc_mask->rq_nxt_blk_addr = 0;
4401 		hr_reg_write(context, QPC_RQ_NXT_BLK_ADDR_H,
4402 			     upper_32_bits(to_hr_hw_page_addr(mtts[1])));
4403 		hr_reg_clear(qpc_mask, QPC_RQ_NXT_BLK_ADDR_H);
4404 	}
4405 
4406 	return 0;
4407 }
4408 
4409 static int config_qp_sq_buf(struct hns_roce_dev *hr_dev,
4410 			    struct hns_roce_qp *hr_qp,
4411 			    struct hns_roce_v2_qp_context *context,
4412 			    struct hns_roce_v2_qp_context *qpc_mask)
4413 {
4414 	struct ib_device *ibdev = &hr_dev->ib_dev;
4415 	u64 sge_cur_blk = 0;
4416 	u64 sq_cur_blk = 0;
4417 	int count;
4418 
4419 	/* search qp buf's mtts */
4420 	count = hns_roce_mtr_find(hr_dev, &hr_qp->mtr, 0, &sq_cur_blk, 1, NULL);
4421 	if (count < 1) {
4422 		ibdev_err(ibdev, "failed to find QP(0x%lx) SQ buf.\n",
4423 			  hr_qp->qpn);
4424 		return -EINVAL;
4425 	}
4426 	if (hr_qp->sge.sge_cnt > 0) {
4427 		count = hns_roce_mtr_find(hr_dev, &hr_qp->mtr,
4428 					  hr_qp->sge.offset,
4429 					  &sge_cur_blk, 1, NULL);
4430 		if (count < 1) {
4431 			ibdev_err(ibdev, "failed to find QP(0x%lx) SGE buf.\n",
4432 				  hr_qp->qpn);
4433 			return -EINVAL;
4434 		}
4435 	}
4436 
4437 	/*
4438 	 * In v2 engine, software pass context and context mask to hardware
4439 	 * when modifying qp. If software need modify some fields in context,
4440 	 * we should set all bits of the relevant fields in context mask to
4441 	 * 0 at the same time, else set them to 0x1.
4442 	 */
4443 	hr_reg_write(context, QPC_SQ_CUR_BLK_ADDR_L,
4444 		     lower_32_bits(to_hr_hw_page_addr(sq_cur_blk)));
4445 	hr_reg_write(context, QPC_SQ_CUR_BLK_ADDR_H,
4446 		     upper_32_bits(to_hr_hw_page_addr(sq_cur_blk)));
4447 	hr_reg_clear(qpc_mask, QPC_SQ_CUR_BLK_ADDR_L);
4448 	hr_reg_clear(qpc_mask, QPC_SQ_CUR_BLK_ADDR_H);
4449 
4450 	hr_reg_write(context, QPC_SQ_CUR_SGE_BLK_ADDR_L,
4451 		     lower_32_bits(to_hr_hw_page_addr(sge_cur_blk)));
4452 	hr_reg_write(context, QPC_SQ_CUR_SGE_BLK_ADDR_H,
4453 		     upper_32_bits(to_hr_hw_page_addr(sge_cur_blk)));
4454 	hr_reg_clear(qpc_mask, QPC_SQ_CUR_SGE_BLK_ADDR_L);
4455 	hr_reg_clear(qpc_mask, QPC_SQ_CUR_SGE_BLK_ADDR_H);
4456 
4457 	hr_reg_write(context, QPC_RX_SQ_CUR_BLK_ADDR_L,
4458 		     lower_32_bits(to_hr_hw_page_addr(sq_cur_blk)));
4459 	hr_reg_write(context, QPC_RX_SQ_CUR_BLK_ADDR_H,
4460 		     upper_32_bits(to_hr_hw_page_addr(sq_cur_blk)));
4461 	hr_reg_clear(qpc_mask, QPC_RX_SQ_CUR_BLK_ADDR_L);
4462 	hr_reg_clear(qpc_mask, QPC_RX_SQ_CUR_BLK_ADDR_H);
4463 
4464 	return 0;
4465 }
4466 
4467 static inline enum ib_mtu get_mtu(struct ib_qp *ibqp,
4468 				  const struct ib_qp_attr *attr)
4469 {
4470 	if (ibqp->qp_type == IB_QPT_GSI || ibqp->qp_type == IB_QPT_UD)
4471 		return IB_MTU_4096;
4472 
4473 	return attr->path_mtu;
4474 }
4475 
4476 static int modify_qp_init_to_rtr(struct ib_qp *ibqp,
4477 				 const struct ib_qp_attr *attr, int attr_mask,
4478 				 struct hns_roce_v2_qp_context *context,
4479 				 struct hns_roce_v2_qp_context *qpc_mask,
4480 				 struct ib_udata *udata)
4481 {
4482 	struct hns_roce_ucontext *uctx = rdma_udata_to_drv_context(udata,
4483 					  struct hns_roce_ucontext, ibucontext);
4484 	struct hns_roce_dev *hr_dev = to_hr_dev(ibqp->device);
4485 	struct hns_roce_qp *hr_qp = to_hr_qp(ibqp);
4486 	struct ib_device *ibdev = &hr_dev->ib_dev;
4487 	dma_addr_t trrl_ba;
4488 	dma_addr_t irrl_ba;
4489 	enum ib_mtu ib_mtu;
4490 	const u8 *smac;
4491 	u8 lp_pktn_ini;
4492 	u64 *mtts;
4493 	u8 *dmac;
4494 	u32 port;
4495 	int mtu;
4496 	int ret;
4497 
4498 	ret = config_qp_rq_buf(hr_dev, hr_qp, context, qpc_mask);
4499 	if (ret) {
4500 		ibdev_err(ibdev, "failed to config rq buf, ret = %d.\n", ret);
4501 		return ret;
4502 	}
4503 
4504 	/* Search IRRL's mtts */
4505 	mtts = hns_roce_table_find(hr_dev, &hr_dev->qp_table.irrl_table,
4506 				   hr_qp->qpn, &irrl_ba);
4507 	if (!mtts) {
4508 		ibdev_err(ibdev, "failed to find qp irrl_table.\n");
4509 		return -EINVAL;
4510 	}
4511 
4512 	/* Search TRRL's mtts */
4513 	mtts = hns_roce_table_find(hr_dev, &hr_dev->qp_table.trrl_table,
4514 				   hr_qp->qpn, &trrl_ba);
4515 	if (!mtts) {
4516 		ibdev_err(ibdev, "failed to find qp trrl_table.\n");
4517 		return -EINVAL;
4518 	}
4519 
4520 	if (attr_mask & IB_QP_ALT_PATH) {
4521 		ibdev_err(ibdev, "INIT2RTR attr_mask (0x%x) error.\n",
4522 			  attr_mask);
4523 		return -EINVAL;
4524 	}
4525 
4526 	hr_reg_write(context, QPC_TRRL_BA_L, trrl_ba >> 4);
4527 	hr_reg_clear(qpc_mask, QPC_TRRL_BA_L);
4528 	context->trrl_ba = cpu_to_le32(trrl_ba >> (16 + 4));
4529 	qpc_mask->trrl_ba = 0;
4530 	hr_reg_write(context, QPC_TRRL_BA_H, trrl_ba >> (32 + 16 + 4));
4531 	hr_reg_clear(qpc_mask, QPC_TRRL_BA_H);
4532 
4533 	context->irrl_ba = cpu_to_le32(irrl_ba >> 6);
4534 	qpc_mask->irrl_ba = 0;
4535 	hr_reg_write(context, QPC_IRRL_BA_H, irrl_ba >> (32 + 6));
4536 	hr_reg_clear(qpc_mask, QPC_IRRL_BA_H);
4537 
4538 	hr_reg_enable(context, QPC_RMT_E2E);
4539 	hr_reg_clear(qpc_mask, QPC_RMT_E2E);
4540 
4541 	hr_reg_write(context, QPC_SIG_TYPE, hr_qp->sq_signal_bits);
4542 	hr_reg_clear(qpc_mask, QPC_SIG_TYPE);
4543 
4544 	port = (attr_mask & IB_QP_PORT) ? (attr->port_num - 1) : hr_qp->port;
4545 
4546 	smac = (const u8 *)hr_dev->dev_addr[port];
4547 	dmac = (u8 *)attr->ah_attr.roce.dmac;
4548 	/* when dmac equals smac or loop_idc is 1, it should loopback */
4549 	if (ether_addr_equal_unaligned(dmac, smac) ||
4550 	    hr_dev->loop_idc == 0x1) {
4551 		hr_reg_write(context, QPC_LBI, hr_dev->loop_idc);
4552 		hr_reg_clear(qpc_mask, QPC_LBI);
4553 	}
4554 
4555 	if (attr_mask & IB_QP_DEST_QPN) {
4556 		hr_reg_write(context, QPC_DQPN, attr->dest_qp_num);
4557 		hr_reg_clear(qpc_mask, QPC_DQPN);
4558 	}
4559 
4560 	memcpy(&context->dmac, dmac, sizeof(u32));
4561 	hr_reg_write(context, QPC_DMAC_H, *((u16 *)(&dmac[4])));
4562 	qpc_mask->dmac = 0;
4563 	hr_reg_clear(qpc_mask, QPC_DMAC_H);
4564 
4565 	ib_mtu = get_mtu(ibqp, attr);
4566 	hr_qp->path_mtu = ib_mtu;
4567 
4568 	mtu = ib_mtu_enum_to_int(ib_mtu);
4569 	if (WARN_ON(mtu <= 0))
4570 		return -EINVAL;
4571 #define MIN_LP_MSG_LEN 1024
4572 	/* mtu * (2 ^ lp_pktn_ini) should be in the range of 1024 to mtu */
4573 	lp_pktn_ini = ilog2(max(mtu, MIN_LP_MSG_LEN) / mtu);
4574 
4575 	if (attr_mask & IB_QP_PATH_MTU) {
4576 		hr_reg_write(context, QPC_MTU, ib_mtu);
4577 		hr_reg_clear(qpc_mask, QPC_MTU);
4578 	}
4579 
4580 	hr_reg_write(context, QPC_LP_PKTN_INI, lp_pktn_ini);
4581 	hr_reg_clear(qpc_mask, QPC_LP_PKTN_INI);
4582 
4583 	/* ACK_REQ_FREQ should be larger than or equal to LP_PKTN_INI */
4584 	hr_reg_write(context, QPC_ACK_REQ_FREQ, lp_pktn_ini);
4585 	hr_reg_clear(qpc_mask, QPC_ACK_REQ_FREQ);
4586 
4587 	hr_reg_clear(qpc_mask, QPC_RX_REQ_PSN_ERR);
4588 	hr_reg_clear(qpc_mask, QPC_RX_REQ_MSN);
4589 	hr_reg_clear(qpc_mask, QPC_RX_REQ_LAST_OPTYPE);
4590 
4591 	context->rq_rnr_timer = 0;
4592 	qpc_mask->rq_rnr_timer = 0;
4593 
4594 	hr_reg_clear(qpc_mask, QPC_TRRL_HEAD_MAX);
4595 	hr_reg_clear(qpc_mask, QPC_TRRL_TAIL_MAX);
4596 
4597 	/* rocee send 2^lp_sgen_ini segs every time */
4598 	hr_reg_write(context, QPC_LP_SGEN_INI, 3);
4599 	hr_reg_clear(qpc_mask, QPC_LP_SGEN_INI);
4600 
4601 	if (udata && ibqp->qp_type == IB_QPT_RC &&
4602 	    (uctx->config & HNS_ROCE_RQ_INLINE_FLAGS)) {
4603 		hr_reg_write_bool(context, QPC_RQIE,
4604 				  hr_dev->caps.flags &
4605 				  HNS_ROCE_CAP_FLAG_RQ_INLINE);
4606 		hr_reg_clear(qpc_mask, QPC_RQIE);
4607 	}
4608 
4609 	if (udata &&
4610 	    (ibqp->qp_type == IB_QPT_RC || ibqp->qp_type == IB_QPT_XRC_TGT) &&
4611 	    (uctx->config & HNS_ROCE_CQE_INLINE_FLAGS)) {
4612 		hr_reg_write_bool(context, QPC_CQEIE,
4613 				  hr_dev->caps.flags &
4614 				  HNS_ROCE_CAP_FLAG_CQE_INLINE);
4615 		hr_reg_clear(qpc_mask, QPC_CQEIE);
4616 
4617 		hr_reg_write(context, QPC_CQEIS, 0);
4618 		hr_reg_clear(qpc_mask, QPC_CQEIS);
4619 	}
4620 
4621 	return 0;
4622 }
4623 
4624 static int modify_qp_rtr_to_rts(struct ib_qp *ibqp,
4625 				const struct ib_qp_attr *attr, int attr_mask,
4626 				struct hns_roce_v2_qp_context *context,
4627 				struct hns_roce_v2_qp_context *qpc_mask)
4628 {
4629 	struct hns_roce_dev *hr_dev = to_hr_dev(ibqp->device);
4630 	struct hns_roce_qp *hr_qp = to_hr_qp(ibqp);
4631 	struct ib_device *ibdev = &hr_dev->ib_dev;
4632 	int ret;
4633 
4634 	/* Not support alternate path and path migration */
4635 	if (attr_mask & (IB_QP_ALT_PATH | IB_QP_PATH_MIG_STATE)) {
4636 		ibdev_err(ibdev, "RTR2RTS attr_mask (0x%x)error\n", attr_mask);
4637 		return -EINVAL;
4638 	}
4639 
4640 	ret = config_qp_sq_buf(hr_dev, hr_qp, context, qpc_mask);
4641 	if (ret) {
4642 		ibdev_err(ibdev, "failed to config sq buf, ret = %d.\n", ret);
4643 		return ret;
4644 	}
4645 
4646 	/*
4647 	 * Set some fields in context to zero, Because the default values
4648 	 * of all fields in context are zero, we need not set them to 0 again.
4649 	 * but we should set the relevant fields of context mask to 0.
4650 	 */
4651 	hr_reg_clear(qpc_mask, QPC_IRRL_SGE_IDX);
4652 
4653 	hr_reg_clear(qpc_mask, QPC_RX_ACK_MSN);
4654 
4655 	hr_reg_clear(qpc_mask, QPC_ACK_LAST_OPTYPE);
4656 	hr_reg_clear(qpc_mask, QPC_IRRL_PSN_VLD);
4657 	hr_reg_clear(qpc_mask, QPC_IRRL_PSN);
4658 
4659 	hr_reg_clear(qpc_mask, QPC_IRRL_TAIL_REAL);
4660 
4661 	hr_reg_clear(qpc_mask, QPC_RETRY_MSG_MSN);
4662 
4663 	hr_reg_clear(qpc_mask, QPC_RNR_RETRY_FLAG);
4664 
4665 	hr_reg_clear(qpc_mask, QPC_CHECK_FLG);
4666 
4667 	hr_reg_clear(qpc_mask, QPC_V2_IRRL_HEAD);
4668 
4669 	return 0;
4670 }
4671 
4672 static int get_dip_ctx_idx(struct ib_qp *ibqp, const struct ib_qp_attr *attr,
4673 			   u32 *dip_idx)
4674 {
4675 	const struct ib_global_route *grh = rdma_ah_read_grh(&attr->ah_attr);
4676 	struct hns_roce_dev *hr_dev = to_hr_dev(ibqp->device);
4677 	u32 *spare_idx = hr_dev->qp_table.idx_table.spare_idx;
4678 	u32 *head =  &hr_dev->qp_table.idx_table.head;
4679 	u32 *tail =  &hr_dev->qp_table.idx_table.tail;
4680 	struct hns_roce_dip *hr_dip;
4681 	unsigned long flags;
4682 	int ret = 0;
4683 
4684 	spin_lock_irqsave(&hr_dev->dip_list_lock, flags);
4685 
4686 	spare_idx[*tail] = ibqp->qp_num;
4687 	*tail = (*tail == hr_dev->caps.num_qps - 1) ? 0 : (*tail + 1);
4688 
4689 	list_for_each_entry(hr_dip, &hr_dev->dip_list, node) {
4690 		if (!memcmp(grh->dgid.raw, hr_dip->dgid, 16)) {
4691 			*dip_idx = hr_dip->dip_idx;
4692 			goto out;
4693 		}
4694 	}
4695 
4696 	/* If no dgid is found, a new dip and a mapping between dgid and
4697 	 * dip_idx will be created.
4698 	 */
4699 	hr_dip = kzalloc(sizeof(*hr_dip), GFP_ATOMIC);
4700 	if (!hr_dip) {
4701 		ret = -ENOMEM;
4702 		goto out;
4703 	}
4704 
4705 	memcpy(hr_dip->dgid, grh->dgid.raw, sizeof(grh->dgid.raw));
4706 	hr_dip->dip_idx = *dip_idx = spare_idx[*head];
4707 	*head = (*head == hr_dev->caps.num_qps - 1) ? 0 : (*head + 1);
4708 	list_add_tail(&hr_dip->node, &hr_dev->dip_list);
4709 
4710 out:
4711 	spin_unlock_irqrestore(&hr_dev->dip_list_lock, flags);
4712 	return ret;
4713 }
4714 
4715 enum {
4716 	CONG_DCQCN,
4717 	CONG_WINDOW,
4718 };
4719 
4720 enum {
4721 	UNSUPPORT_CONG_LEVEL,
4722 	SUPPORT_CONG_LEVEL,
4723 };
4724 
4725 enum {
4726 	CONG_LDCP,
4727 	CONG_HC3,
4728 };
4729 
4730 enum {
4731 	DIP_INVALID,
4732 	DIP_VALID,
4733 };
4734 
4735 enum {
4736 	WND_LIMIT,
4737 	WND_UNLIMIT,
4738 };
4739 
4740 static int check_cong_type(struct ib_qp *ibqp,
4741 			   struct hns_roce_congestion_algorithm *cong_alg)
4742 {
4743 	struct hns_roce_dev *hr_dev = to_hr_dev(ibqp->device);
4744 	struct hns_roce_qp *hr_qp = to_hr_qp(ibqp);
4745 
4746 	if (ibqp->qp_type == IB_QPT_UD || ibqp->qp_type == IB_QPT_GSI)
4747 		hr_qp->cong_type = CONG_TYPE_DCQCN;
4748 	else
4749 		hr_qp->cong_type = hr_dev->caps.cong_type;
4750 
4751 	/* different congestion types match different configurations */
4752 	switch (hr_qp->cong_type) {
4753 	case CONG_TYPE_DCQCN:
4754 		cong_alg->alg_sel = CONG_DCQCN;
4755 		cong_alg->alg_sub_sel = UNSUPPORT_CONG_LEVEL;
4756 		cong_alg->dip_vld = DIP_INVALID;
4757 		cong_alg->wnd_mode_sel = WND_LIMIT;
4758 		break;
4759 	case CONG_TYPE_LDCP:
4760 		cong_alg->alg_sel = CONG_WINDOW;
4761 		cong_alg->alg_sub_sel = CONG_LDCP;
4762 		cong_alg->dip_vld = DIP_INVALID;
4763 		cong_alg->wnd_mode_sel = WND_UNLIMIT;
4764 		break;
4765 	case CONG_TYPE_HC3:
4766 		cong_alg->alg_sel = CONG_WINDOW;
4767 		cong_alg->alg_sub_sel = CONG_HC3;
4768 		cong_alg->dip_vld = DIP_INVALID;
4769 		cong_alg->wnd_mode_sel = WND_LIMIT;
4770 		break;
4771 	case CONG_TYPE_DIP:
4772 		cong_alg->alg_sel = CONG_DCQCN;
4773 		cong_alg->alg_sub_sel = UNSUPPORT_CONG_LEVEL;
4774 		cong_alg->dip_vld = DIP_VALID;
4775 		cong_alg->wnd_mode_sel = WND_LIMIT;
4776 		break;
4777 	default:
4778 		ibdev_warn(&hr_dev->ib_dev,
4779 			   "invalid type(%u) for congestion selection.\n",
4780 			   hr_qp->cong_type);
4781 		hr_qp->cong_type = CONG_TYPE_DCQCN;
4782 		cong_alg->alg_sel = CONG_DCQCN;
4783 		cong_alg->alg_sub_sel = UNSUPPORT_CONG_LEVEL;
4784 		cong_alg->dip_vld = DIP_INVALID;
4785 		cong_alg->wnd_mode_sel = WND_LIMIT;
4786 		break;
4787 	}
4788 
4789 	return 0;
4790 }
4791 
4792 static int fill_cong_field(struct ib_qp *ibqp, const struct ib_qp_attr *attr,
4793 			   struct hns_roce_v2_qp_context *context,
4794 			   struct hns_roce_v2_qp_context *qpc_mask)
4795 {
4796 	const struct ib_global_route *grh = rdma_ah_read_grh(&attr->ah_attr);
4797 	struct hns_roce_congestion_algorithm cong_field;
4798 	struct ib_device *ibdev = ibqp->device;
4799 	struct hns_roce_dev *hr_dev = to_hr_dev(ibdev);
4800 	struct hns_roce_qp *hr_qp = to_hr_qp(ibqp);
4801 	u32 dip_idx = 0;
4802 	int ret;
4803 
4804 	if (hr_dev->pci_dev->revision == PCI_REVISION_ID_HIP08 ||
4805 	    grh->sgid_attr->gid_type == IB_GID_TYPE_ROCE)
4806 		return 0;
4807 
4808 	ret = check_cong_type(ibqp, &cong_field);
4809 	if (ret)
4810 		return ret;
4811 
4812 	hr_reg_write(context, QPC_CONG_ALGO_TMPL_ID, hr_dev->cong_algo_tmpl_id +
4813 		     hr_qp->cong_type * HNS_ROCE_CONG_SIZE);
4814 	hr_reg_clear(qpc_mask, QPC_CONG_ALGO_TMPL_ID);
4815 	hr_reg_write(&context->ext, QPCEX_CONG_ALG_SEL, cong_field.alg_sel);
4816 	hr_reg_clear(&qpc_mask->ext, QPCEX_CONG_ALG_SEL);
4817 	hr_reg_write(&context->ext, QPCEX_CONG_ALG_SUB_SEL,
4818 		     cong_field.alg_sub_sel);
4819 	hr_reg_clear(&qpc_mask->ext, QPCEX_CONG_ALG_SUB_SEL);
4820 	hr_reg_write(&context->ext, QPCEX_DIP_CTX_IDX_VLD, cong_field.dip_vld);
4821 	hr_reg_clear(&qpc_mask->ext, QPCEX_DIP_CTX_IDX_VLD);
4822 	hr_reg_write(&context->ext, QPCEX_SQ_RQ_NOT_FORBID_EN,
4823 		     cong_field.wnd_mode_sel);
4824 	hr_reg_clear(&qpc_mask->ext, QPCEX_SQ_RQ_NOT_FORBID_EN);
4825 
4826 	/* if dip is disabled, there is no need to set dip idx */
4827 	if (cong_field.dip_vld == 0)
4828 		return 0;
4829 
4830 	ret = get_dip_ctx_idx(ibqp, attr, &dip_idx);
4831 	if (ret) {
4832 		ibdev_err(ibdev, "failed to fill cong field, ret = %d.\n", ret);
4833 		return ret;
4834 	}
4835 
4836 	hr_reg_write(&context->ext, QPCEX_DIP_CTX_IDX, dip_idx);
4837 	hr_reg_write(&qpc_mask->ext, QPCEX_DIP_CTX_IDX, 0);
4838 
4839 	return 0;
4840 }
4841 
4842 static int hns_roce_v2_set_path(struct ib_qp *ibqp,
4843 				const struct ib_qp_attr *attr,
4844 				int attr_mask,
4845 				struct hns_roce_v2_qp_context *context,
4846 				struct hns_roce_v2_qp_context *qpc_mask)
4847 {
4848 	const struct ib_global_route *grh = rdma_ah_read_grh(&attr->ah_attr);
4849 	struct hns_roce_dev *hr_dev = to_hr_dev(ibqp->device);
4850 	struct hns_roce_qp *hr_qp = to_hr_qp(ibqp);
4851 	struct ib_device *ibdev = &hr_dev->ib_dev;
4852 	const struct ib_gid_attr *gid_attr = NULL;
4853 	u8 sl = rdma_ah_get_sl(&attr->ah_attr);
4854 	int is_roce_protocol;
4855 	u16 vlan_id = 0xffff;
4856 	bool is_udp = false;
4857 	u32 max_sl;
4858 	u8 ib_port;
4859 	u8 hr_port;
4860 	int ret;
4861 
4862 	max_sl = min_t(u32, MAX_SERVICE_LEVEL, hr_dev->caps.sl_num - 1);
4863 	if (unlikely(sl > max_sl)) {
4864 		ibdev_err_ratelimited(ibdev,
4865 				      "failed to fill QPC, sl (%u) shouldn't be larger than %u.\n",
4866 				      sl, max_sl);
4867 		return -EINVAL;
4868 	}
4869 
4870 	/*
4871 	 * If free_mr_en of qp is set, it means that this qp comes from
4872 	 * free mr. This qp will perform the loopback operation.
4873 	 * In the loopback scenario, only sl needs to be set.
4874 	 */
4875 	if (hr_qp->free_mr_en) {
4876 		hr_reg_write(context, QPC_SL, sl);
4877 		hr_reg_clear(qpc_mask, QPC_SL);
4878 		hr_qp->sl = sl;
4879 		return 0;
4880 	}
4881 
4882 	ib_port = (attr_mask & IB_QP_PORT) ? attr->port_num : hr_qp->port + 1;
4883 	hr_port = ib_port - 1;
4884 	is_roce_protocol = rdma_cap_eth_ah(&hr_dev->ib_dev, ib_port) &&
4885 			   rdma_ah_get_ah_flags(&attr->ah_attr) & IB_AH_GRH;
4886 
4887 	if (is_roce_protocol) {
4888 		gid_attr = attr->ah_attr.grh.sgid_attr;
4889 		ret = rdma_read_gid_l2_fields(gid_attr, &vlan_id, NULL);
4890 		if (ret)
4891 			return ret;
4892 
4893 		is_udp = (gid_attr->gid_type == IB_GID_TYPE_ROCE_UDP_ENCAP);
4894 	}
4895 
4896 	/* Only HIP08 needs to set the vlan_en bits in QPC */
4897 	if (vlan_id < VLAN_N_VID &&
4898 	    hr_dev->pci_dev->revision == PCI_REVISION_ID_HIP08) {
4899 		hr_reg_enable(context, QPC_RQ_VLAN_EN);
4900 		hr_reg_clear(qpc_mask, QPC_RQ_VLAN_EN);
4901 		hr_reg_enable(context, QPC_SQ_VLAN_EN);
4902 		hr_reg_clear(qpc_mask, QPC_SQ_VLAN_EN);
4903 	}
4904 
4905 	hr_reg_write(context, QPC_VLAN_ID, vlan_id);
4906 	hr_reg_clear(qpc_mask, QPC_VLAN_ID);
4907 
4908 	if (grh->sgid_index >= hr_dev->caps.gid_table_len[hr_port]) {
4909 		ibdev_err(ibdev, "sgid_index(%u) too large. max is %d\n",
4910 			  grh->sgid_index, hr_dev->caps.gid_table_len[hr_port]);
4911 		return -EINVAL;
4912 	}
4913 
4914 	if (attr->ah_attr.type != RDMA_AH_ATTR_TYPE_ROCE) {
4915 		ibdev_err(ibdev, "ah attr is not RDMA roce type\n");
4916 		return -EINVAL;
4917 	}
4918 
4919 	hr_reg_write(context, QPC_UDPSPN,
4920 		     is_udp ? rdma_get_udp_sport(grh->flow_label, ibqp->qp_num,
4921 						 attr->dest_qp_num) :
4922 				    0);
4923 
4924 	hr_reg_clear(qpc_mask, QPC_UDPSPN);
4925 
4926 	hr_reg_write(context, QPC_GMV_IDX, grh->sgid_index);
4927 
4928 	hr_reg_clear(qpc_mask, QPC_GMV_IDX);
4929 
4930 	hr_reg_write(context, QPC_HOPLIMIT, grh->hop_limit);
4931 	hr_reg_clear(qpc_mask, QPC_HOPLIMIT);
4932 
4933 	ret = fill_cong_field(ibqp, attr, context, qpc_mask);
4934 	if (ret)
4935 		return ret;
4936 
4937 	hr_reg_write(context, QPC_TC, get_tclass(&attr->ah_attr.grh));
4938 	hr_reg_clear(qpc_mask, QPC_TC);
4939 
4940 	hr_reg_write(context, QPC_FL, grh->flow_label);
4941 	hr_reg_clear(qpc_mask, QPC_FL);
4942 	memcpy(context->dgid, grh->dgid.raw, sizeof(grh->dgid.raw));
4943 	memset(qpc_mask->dgid, 0, sizeof(grh->dgid.raw));
4944 
4945 	hr_qp->sl = sl;
4946 	hr_reg_write(context, QPC_SL, hr_qp->sl);
4947 	hr_reg_clear(qpc_mask, QPC_SL);
4948 
4949 	return 0;
4950 }
4951 
4952 static bool check_qp_state(enum ib_qp_state cur_state,
4953 			   enum ib_qp_state new_state)
4954 {
4955 	static const bool sm[][IB_QPS_ERR + 1] = {
4956 		[IB_QPS_RESET] = { [IB_QPS_RESET] = true,
4957 				   [IB_QPS_INIT] = true },
4958 		[IB_QPS_INIT] = { [IB_QPS_RESET] = true,
4959 				  [IB_QPS_INIT] = true,
4960 				  [IB_QPS_RTR] = true,
4961 				  [IB_QPS_ERR] = true },
4962 		[IB_QPS_RTR] = { [IB_QPS_RESET] = true,
4963 				 [IB_QPS_RTS] = true,
4964 				 [IB_QPS_ERR] = true },
4965 		[IB_QPS_RTS] = { [IB_QPS_RESET] = true,
4966 				 [IB_QPS_RTS] = true,
4967 				 [IB_QPS_ERR] = true },
4968 		[IB_QPS_SQD] = {},
4969 		[IB_QPS_SQE] = {},
4970 		[IB_QPS_ERR] = { [IB_QPS_RESET] = true,
4971 				 [IB_QPS_ERR] = true }
4972 	};
4973 
4974 	return sm[cur_state][new_state];
4975 }
4976 
4977 static int hns_roce_v2_set_abs_fields(struct ib_qp *ibqp,
4978 				      const struct ib_qp_attr *attr,
4979 				      int attr_mask,
4980 				      enum ib_qp_state cur_state,
4981 				      enum ib_qp_state new_state,
4982 				      struct hns_roce_v2_qp_context *context,
4983 				      struct hns_roce_v2_qp_context *qpc_mask,
4984 				      struct ib_udata *udata)
4985 {
4986 	struct hns_roce_dev *hr_dev = to_hr_dev(ibqp->device);
4987 	int ret = 0;
4988 
4989 	if (!check_qp_state(cur_state, new_state)) {
4990 		ibdev_err(&hr_dev->ib_dev, "Illegal state for QP!\n");
4991 		return -EINVAL;
4992 	}
4993 
4994 	if (cur_state == IB_QPS_RESET && new_state == IB_QPS_INIT) {
4995 		memset(qpc_mask, 0, hr_dev->caps.qpc_sz);
4996 		modify_qp_reset_to_init(ibqp, attr, context, qpc_mask);
4997 	} else if (cur_state == IB_QPS_INIT && new_state == IB_QPS_INIT) {
4998 		modify_qp_init_to_init(ibqp, attr, context, qpc_mask);
4999 	} else if (cur_state == IB_QPS_INIT && new_state == IB_QPS_RTR) {
5000 		ret = modify_qp_init_to_rtr(ibqp, attr, attr_mask, context,
5001 					    qpc_mask, udata);
5002 	} else if (cur_state == IB_QPS_RTR && new_state == IB_QPS_RTS) {
5003 		ret = modify_qp_rtr_to_rts(ibqp, attr, attr_mask, context,
5004 					   qpc_mask);
5005 	}
5006 
5007 	return ret;
5008 }
5009 
5010 static bool check_qp_timeout_cfg_range(struct hns_roce_dev *hr_dev, u8 *timeout)
5011 {
5012 #define QP_ACK_TIMEOUT_MAX_HIP08 20
5013 #define QP_ACK_TIMEOUT_MAX 31
5014 
5015 	if (hr_dev->pci_dev->revision == PCI_REVISION_ID_HIP08) {
5016 		if (*timeout > QP_ACK_TIMEOUT_MAX_HIP08) {
5017 			ibdev_warn(&hr_dev->ib_dev,
5018 				   "local ACK timeout shall be 0 to 20.\n");
5019 			return false;
5020 		}
5021 		*timeout += HNS_ROCE_V2_QP_ACK_TIMEOUT_OFS_HIP08;
5022 	} else if (hr_dev->pci_dev->revision > PCI_REVISION_ID_HIP08) {
5023 		if (*timeout > QP_ACK_TIMEOUT_MAX) {
5024 			ibdev_warn(&hr_dev->ib_dev,
5025 				   "local ACK timeout shall be 0 to 31.\n");
5026 			return false;
5027 		}
5028 	}
5029 
5030 	return true;
5031 }
5032 
5033 static int hns_roce_v2_set_opt_fields(struct ib_qp *ibqp,
5034 				      const struct ib_qp_attr *attr,
5035 				      int attr_mask,
5036 				      struct hns_roce_v2_qp_context *context,
5037 				      struct hns_roce_v2_qp_context *qpc_mask)
5038 {
5039 	struct hns_roce_dev *hr_dev = to_hr_dev(ibqp->device);
5040 	struct hns_roce_qp *hr_qp = to_hr_qp(ibqp);
5041 	int ret = 0;
5042 	u8 timeout;
5043 
5044 	if (attr_mask & IB_QP_AV) {
5045 		ret = hns_roce_v2_set_path(ibqp, attr, attr_mask, context,
5046 					   qpc_mask);
5047 		if (ret)
5048 			return ret;
5049 	}
5050 
5051 	if (attr_mask & IB_QP_TIMEOUT) {
5052 		timeout = attr->timeout;
5053 		if (check_qp_timeout_cfg_range(hr_dev, &timeout)) {
5054 			hr_reg_write(context, QPC_AT, timeout);
5055 			hr_reg_clear(qpc_mask, QPC_AT);
5056 		}
5057 	}
5058 
5059 	if (attr_mask & IB_QP_RETRY_CNT) {
5060 		hr_reg_write(context, QPC_RETRY_NUM_INIT, attr->retry_cnt);
5061 		hr_reg_clear(qpc_mask, QPC_RETRY_NUM_INIT);
5062 
5063 		hr_reg_write(context, QPC_RETRY_CNT, attr->retry_cnt);
5064 		hr_reg_clear(qpc_mask, QPC_RETRY_CNT);
5065 	}
5066 
5067 	if (attr_mask & IB_QP_RNR_RETRY) {
5068 		hr_reg_write(context, QPC_RNR_NUM_INIT, attr->rnr_retry);
5069 		hr_reg_clear(qpc_mask, QPC_RNR_NUM_INIT);
5070 
5071 		hr_reg_write(context, QPC_RNR_CNT, attr->rnr_retry);
5072 		hr_reg_clear(qpc_mask, QPC_RNR_CNT);
5073 	}
5074 
5075 	if (attr_mask & IB_QP_SQ_PSN) {
5076 		hr_reg_write(context, QPC_SQ_CUR_PSN, attr->sq_psn);
5077 		hr_reg_clear(qpc_mask, QPC_SQ_CUR_PSN);
5078 
5079 		hr_reg_write(context, QPC_SQ_MAX_PSN, attr->sq_psn);
5080 		hr_reg_clear(qpc_mask, QPC_SQ_MAX_PSN);
5081 
5082 		hr_reg_write(context, QPC_RETRY_MSG_PSN_L, attr->sq_psn);
5083 		hr_reg_clear(qpc_mask, QPC_RETRY_MSG_PSN_L);
5084 
5085 		hr_reg_write(context, QPC_RETRY_MSG_PSN_H,
5086 			     attr->sq_psn >> RETRY_MSG_PSN_SHIFT);
5087 		hr_reg_clear(qpc_mask, QPC_RETRY_MSG_PSN_H);
5088 
5089 		hr_reg_write(context, QPC_RETRY_MSG_FPKT_PSN, attr->sq_psn);
5090 		hr_reg_clear(qpc_mask, QPC_RETRY_MSG_FPKT_PSN);
5091 
5092 		hr_reg_write(context, QPC_RX_ACK_EPSN, attr->sq_psn);
5093 		hr_reg_clear(qpc_mask, QPC_RX_ACK_EPSN);
5094 	}
5095 
5096 	if ((attr_mask & IB_QP_MAX_DEST_RD_ATOMIC) &&
5097 	     attr->max_dest_rd_atomic) {
5098 		hr_reg_write(context, QPC_RR_MAX,
5099 			     fls(attr->max_dest_rd_atomic - 1));
5100 		hr_reg_clear(qpc_mask, QPC_RR_MAX);
5101 	}
5102 
5103 	if ((attr_mask & IB_QP_MAX_QP_RD_ATOMIC) && attr->max_rd_atomic) {
5104 		hr_reg_write(context, QPC_SR_MAX, fls(attr->max_rd_atomic - 1));
5105 		hr_reg_clear(qpc_mask, QPC_SR_MAX);
5106 	}
5107 
5108 	if (attr_mask & (IB_QP_ACCESS_FLAGS | IB_QP_MAX_DEST_RD_ATOMIC))
5109 		set_access_flags(hr_qp, context, qpc_mask, attr, attr_mask);
5110 
5111 	if (attr_mask & IB_QP_MIN_RNR_TIMER) {
5112 		hr_reg_write(context, QPC_MIN_RNR_TIME,
5113 			    hr_dev->pci_dev->revision == PCI_REVISION_ID_HIP08 ?
5114 			    HNS_ROCE_RNR_TIMER_10NS : attr->min_rnr_timer);
5115 		hr_reg_clear(qpc_mask, QPC_MIN_RNR_TIME);
5116 	}
5117 
5118 	if (attr_mask & IB_QP_RQ_PSN) {
5119 		hr_reg_write(context, QPC_RX_REQ_EPSN, attr->rq_psn);
5120 		hr_reg_clear(qpc_mask, QPC_RX_REQ_EPSN);
5121 
5122 		hr_reg_write(context, QPC_RAQ_PSN, attr->rq_psn - 1);
5123 		hr_reg_clear(qpc_mask, QPC_RAQ_PSN);
5124 	}
5125 
5126 	if (attr_mask & IB_QP_QKEY) {
5127 		context->qkey_xrcd = cpu_to_le32(attr->qkey);
5128 		qpc_mask->qkey_xrcd = 0;
5129 		hr_qp->qkey = attr->qkey;
5130 	}
5131 
5132 	return ret;
5133 }
5134 
5135 static void hns_roce_v2_record_opt_fields(struct ib_qp *ibqp,
5136 					  const struct ib_qp_attr *attr,
5137 					  int attr_mask)
5138 {
5139 	struct hns_roce_dev *hr_dev = to_hr_dev(ibqp->device);
5140 	struct hns_roce_qp *hr_qp = to_hr_qp(ibqp);
5141 
5142 	if (attr_mask & IB_QP_ACCESS_FLAGS)
5143 		hr_qp->atomic_rd_en = attr->qp_access_flags;
5144 
5145 	if (attr_mask & IB_QP_MAX_DEST_RD_ATOMIC)
5146 		hr_qp->resp_depth = attr->max_dest_rd_atomic;
5147 	if (attr_mask & IB_QP_PORT) {
5148 		hr_qp->port = attr->port_num - 1;
5149 		hr_qp->phy_port = hr_dev->iboe.phy_port[hr_qp->port];
5150 	}
5151 }
5152 
5153 static void clear_qp(struct hns_roce_qp *hr_qp)
5154 {
5155 	struct ib_qp *ibqp = &hr_qp->ibqp;
5156 
5157 	if (ibqp->send_cq)
5158 		hns_roce_v2_cq_clean(to_hr_cq(ibqp->send_cq),
5159 				     hr_qp->qpn, NULL);
5160 
5161 	if (ibqp->recv_cq  && ibqp->recv_cq != ibqp->send_cq)
5162 		hns_roce_v2_cq_clean(to_hr_cq(ibqp->recv_cq),
5163 				     hr_qp->qpn, ibqp->srq ?
5164 				     to_hr_srq(ibqp->srq) : NULL);
5165 
5166 	if (hr_qp->en_flags & HNS_ROCE_QP_CAP_RQ_RECORD_DB)
5167 		*hr_qp->rdb.db_record = 0;
5168 
5169 	hr_qp->rq.head = 0;
5170 	hr_qp->rq.tail = 0;
5171 	hr_qp->sq.head = 0;
5172 	hr_qp->sq.tail = 0;
5173 	hr_qp->next_sge = 0;
5174 }
5175 
5176 static void v2_set_flushed_fields(struct ib_qp *ibqp,
5177 				  struct hns_roce_v2_qp_context *context,
5178 				  struct hns_roce_v2_qp_context *qpc_mask)
5179 {
5180 	struct hns_roce_qp *hr_qp = to_hr_qp(ibqp);
5181 	unsigned long sq_flag = 0;
5182 	unsigned long rq_flag = 0;
5183 
5184 	if (ibqp->qp_type == IB_QPT_XRC_TGT)
5185 		return;
5186 
5187 	spin_lock_irqsave(&hr_qp->sq.lock, sq_flag);
5188 	hr_reg_write(context, QPC_SQ_PRODUCER_IDX, hr_qp->sq.head);
5189 	hr_reg_clear(qpc_mask, QPC_SQ_PRODUCER_IDX);
5190 	hr_qp->state = IB_QPS_ERR;
5191 	spin_unlock_irqrestore(&hr_qp->sq.lock, sq_flag);
5192 
5193 	if (ibqp->srq || ibqp->qp_type == IB_QPT_XRC_INI) /* no RQ */
5194 		return;
5195 
5196 	spin_lock_irqsave(&hr_qp->rq.lock, rq_flag);
5197 	hr_reg_write(context, QPC_RQ_PRODUCER_IDX, hr_qp->rq.head);
5198 	hr_reg_clear(qpc_mask, QPC_RQ_PRODUCER_IDX);
5199 	spin_unlock_irqrestore(&hr_qp->rq.lock, rq_flag);
5200 }
5201 
5202 static int hns_roce_v2_modify_qp(struct ib_qp *ibqp,
5203 				 const struct ib_qp_attr *attr,
5204 				 int attr_mask, enum ib_qp_state cur_state,
5205 				 enum ib_qp_state new_state, struct ib_udata *udata)
5206 {
5207 	struct hns_roce_dev *hr_dev = to_hr_dev(ibqp->device);
5208 	struct hns_roce_qp *hr_qp = to_hr_qp(ibqp);
5209 	struct hns_roce_v2_qp_context ctx[2];
5210 	struct hns_roce_v2_qp_context *context = ctx;
5211 	struct hns_roce_v2_qp_context *qpc_mask = ctx + 1;
5212 	struct ib_device *ibdev = &hr_dev->ib_dev;
5213 	int ret;
5214 
5215 	if (attr_mask & ~IB_QP_ATTR_STANDARD_BITS)
5216 		return -EOPNOTSUPP;
5217 
5218 	/*
5219 	 * In v2 engine, software pass context and context mask to hardware
5220 	 * when modifying qp. If software need modify some fields in context,
5221 	 * we should set all bits of the relevant fields in context mask to
5222 	 * 0 at the same time, else set them to 0x1.
5223 	 */
5224 	memset(context, 0, hr_dev->caps.qpc_sz);
5225 	memset(qpc_mask, 0xff, hr_dev->caps.qpc_sz);
5226 
5227 	ret = hns_roce_v2_set_abs_fields(ibqp, attr, attr_mask, cur_state,
5228 					 new_state, context, qpc_mask, udata);
5229 	if (ret)
5230 		goto out;
5231 
5232 	/* When QP state is err, SQ and RQ WQE should be flushed */
5233 	if (new_state == IB_QPS_ERR)
5234 		v2_set_flushed_fields(ibqp, context, qpc_mask);
5235 
5236 	/* Configure the optional fields */
5237 	ret = hns_roce_v2_set_opt_fields(ibqp, attr, attr_mask, context,
5238 					 qpc_mask);
5239 	if (ret)
5240 		goto out;
5241 
5242 	hr_reg_write_bool(context, QPC_INV_CREDIT,
5243 			  to_hr_qp_type(hr_qp->ibqp.qp_type) == SERV_TYPE_XRC ||
5244 			  ibqp->srq);
5245 	hr_reg_clear(qpc_mask, QPC_INV_CREDIT);
5246 
5247 	/* Every status migrate must change state */
5248 	hr_reg_write(context, QPC_QP_ST, new_state);
5249 	hr_reg_clear(qpc_mask, QPC_QP_ST);
5250 
5251 	/* SW pass context to HW */
5252 	ret = hns_roce_v2_qp_modify(hr_dev, context, qpc_mask, hr_qp);
5253 	if (ret) {
5254 		ibdev_err(ibdev, "failed to modify QP, ret = %d.\n", ret);
5255 		goto out;
5256 	}
5257 
5258 	hr_qp->state = new_state;
5259 
5260 	hns_roce_v2_record_opt_fields(ibqp, attr, attr_mask);
5261 
5262 	if (new_state == IB_QPS_RESET && !ibqp->uobject)
5263 		clear_qp(hr_qp);
5264 
5265 out:
5266 	return ret;
5267 }
5268 
5269 static int to_ib_qp_st(enum hns_roce_v2_qp_state state)
5270 {
5271 	static const enum ib_qp_state map[] = {
5272 		[HNS_ROCE_QP_ST_RST] = IB_QPS_RESET,
5273 		[HNS_ROCE_QP_ST_INIT] = IB_QPS_INIT,
5274 		[HNS_ROCE_QP_ST_RTR] = IB_QPS_RTR,
5275 		[HNS_ROCE_QP_ST_RTS] = IB_QPS_RTS,
5276 		[HNS_ROCE_QP_ST_SQD] = IB_QPS_SQD,
5277 		[HNS_ROCE_QP_ST_SQER] = IB_QPS_SQE,
5278 		[HNS_ROCE_QP_ST_ERR] = IB_QPS_ERR,
5279 		[HNS_ROCE_QP_ST_SQ_DRAINING] = IB_QPS_SQD
5280 	};
5281 
5282 	return (state < ARRAY_SIZE(map)) ? map[state] : -1;
5283 }
5284 
5285 static int hns_roce_v2_query_qpc(struct hns_roce_dev *hr_dev, u32 qpn,
5286 				 void *buffer)
5287 {
5288 	struct hns_roce_cmd_mailbox *mailbox;
5289 	int ret;
5290 
5291 	mailbox = hns_roce_alloc_cmd_mailbox(hr_dev);
5292 	if (IS_ERR(mailbox))
5293 		return PTR_ERR(mailbox);
5294 
5295 	ret = hns_roce_cmd_mbox(hr_dev, 0, mailbox->dma, HNS_ROCE_CMD_QUERY_QPC,
5296 				qpn);
5297 	if (ret)
5298 		goto out;
5299 
5300 	memcpy(buffer, mailbox->buf, hr_dev->caps.qpc_sz);
5301 
5302 out:
5303 	hns_roce_free_cmd_mailbox(hr_dev, mailbox);
5304 	return ret;
5305 }
5306 
5307 static u8 get_qp_timeout_attr(struct hns_roce_dev *hr_dev,
5308 			      struct hns_roce_v2_qp_context *context)
5309 {
5310 	u8 timeout;
5311 
5312 	timeout = (u8)hr_reg_read(context, QPC_AT);
5313 	if (hr_dev->pci_dev->revision == PCI_REVISION_ID_HIP08)
5314 		timeout -= HNS_ROCE_V2_QP_ACK_TIMEOUT_OFS_HIP08;
5315 
5316 	return timeout;
5317 }
5318 
5319 static int hns_roce_v2_query_qp(struct ib_qp *ibqp, struct ib_qp_attr *qp_attr,
5320 				int qp_attr_mask,
5321 				struct ib_qp_init_attr *qp_init_attr)
5322 {
5323 	struct hns_roce_dev *hr_dev = to_hr_dev(ibqp->device);
5324 	struct hns_roce_qp *hr_qp = to_hr_qp(ibqp);
5325 	struct hns_roce_v2_qp_context context = {};
5326 	struct ib_device *ibdev = &hr_dev->ib_dev;
5327 	int tmp_qp_state;
5328 	int state;
5329 	int ret;
5330 
5331 	memset(qp_attr, 0, sizeof(*qp_attr));
5332 	memset(qp_init_attr, 0, sizeof(*qp_init_attr));
5333 
5334 	mutex_lock(&hr_qp->mutex);
5335 
5336 	if (hr_qp->state == IB_QPS_RESET) {
5337 		qp_attr->qp_state = IB_QPS_RESET;
5338 		ret = 0;
5339 		goto done;
5340 	}
5341 
5342 	ret = hns_roce_v2_query_qpc(hr_dev, hr_qp->qpn, &context);
5343 	if (ret) {
5344 		ibdev_err(ibdev, "failed to query QPC, ret = %d.\n", ret);
5345 		ret = -EINVAL;
5346 		goto out;
5347 	}
5348 
5349 	state = hr_reg_read(&context, QPC_QP_ST);
5350 	tmp_qp_state = to_ib_qp_st((enum hns_roce_v2_qp_state)state);
5351 	if (tmp_qp_state == -1) {
5352 		ibdev_err(ibdev, "Illegal ib_qp_state\n");
5353 		ret = -EINVAL;
5354 		goto out;
5355 	}
5356 	hr_qp->state = (u8)tmp_qp_state;
5357 	qp_attr->qp_state = (enum ib_qp_state)hr_qp->state;
5358 	qp_attr->path_mtu = (enum ib_mtu)hr_reg_read(&context, QPC_MTU);
5359 	qp_attr->path_mig_state = IB_MIG_ARMED;
5360 	qp_attr->ah_attr.type = RDMA_AH_ATTR_TYPE_ROCE;
5361 	if (hr_qp->ibqp.qp_type == IB_QPT_UD)
5362 		qp_attr->qkey = le32_to_cpu(context.qkey_xrcd);
5363 
5364 	qp_attr->rq_psn = hr_reg_read(&context, QPC_RX_REQ_EPSN);
5365 	qp_attr->sq_psn = (u32)hr_reg_read(&context, QPC_SQ_CUR_PSN);
5366 	qp_attr->dest_qp_num = hr_reg_read(&context, QPC_DQPN);
5367 	qp_attr->qp_access_flags =
5368 		((hr_reg_read(&context, QPC_RRE)) << V2_QP_RRE_S) |
5369 		((hr_reg_read(&context, QPC_RWE)) << V2_QP_RWE_S) |
5370 		((hr_reg_read(&context, QPC_ATE)) << V2_QP_ATE_S);
5371 
5372 	if (hr_qp->ibqp.qp_type == IB_QPT_RC ||
5373 	    hr_qp->ibqp.qp_type == IB_QPT_XRC_INI ||
5374 	    hr_qp->ibqp.qp_type == IB_QPT_XRC_TGT) {
5375 		struct ib_global_route *grh =
5376 			rdma_ah_retrieve_grh(&qp_attr->ah_attr);
5377 
5378 		rdma_ah_set_sl(&qp_attr->ah_attr,
5379 			       hr_reg_read(&context, QPC_SL));
5380 		rdma_ah_set_port_num(&qp_attr->ah_attr, hr_qp->port + 1);
5381 		rdma_ah_set_ah_flags(&qp_attr->ah_attr, IB_AH_GRH);
5382 		grh->flow_label = hr_reg_read(&context, QPC_FL);
5383 		grh->sgid_index = hr_reg_read(&context, QPC_GMV_IDX);
5384 		grh->hop_limit = hr_reg_read(&context, QPC_HOPLIMIT);
5385 		grh->traffic_class = hr_reg_read(&context, QPC_TC);
5386 
5387 		memcpy(grh->dgid.raw, context.dgid, sizeof(grh->dgid.raw));
5388 	}
5389 
5390 	qp_attr->port_num = hr_qp->port + 1;
5391 	qp_attr->sq_draining = 0;
5392 	qp_attr->max_rd_atomic = 1 << hr_reg_read(&context, QPC_SR_MAX);
5393 	qp_attr->max_dest_rd_atomic = 1 << hr_reg_read(&context, QPC_RR_MAX);
5394 
5395 	qp_attr->min_rnr_timer = (u8)hr_reg_read(&context, QPC_MIN_RNR_TIME);
5396 	qp_attr->timeout = get_qp_timeout_attr(hr_dev, &context);
5397 	qp_attr->retry_cnt = hr_reg_read(&context, QPC_RETRY_NUM_INIT);
5398 	qp_attr->rnr_retry = hr_reg_read(&context, QPC_RNR_NUM_INIT);
5399 
5400 done:
5401 	qp_attr->cur_qp_state = qp_attr->qp_state;
5402 	qp_attr->cap.max_recv_wr = hr_qp->rq.wqe_cnt;
5403 	qp_attr->cap.max_recv_sge = hr_qp->rq.max_gs - hr_qp->rq.rsv_sge;
5404 	qp_attr->cap.max_inline_data = hr_qp->max_inline_data;
5405 
5406 	qp_attr->cap.max_send_wr = hr_qp->sq.wqe_cnt;
5407 	qp_attr->cap.max_send_sge = hr_qp->sq.max_gs;
5408 
5409 	qp_init_attr->qp_context = ibqp->qp_context;
5410 	qp_init_attr->qp_type = ibqp->qp_type;
5411 	qp_init_attr->recv_cq = ibqp->recv_cq;
5412 	qp_init_attr->send_cq = ibqp->send_cq;
5413 	qp_init_attr->srq = ibqp->srq;
5414 	qp_init_attr->cap = qp_attr->cap;
5415 	qp_init_attr->sq_sig_type = hr_qp->sq_signal_bits;
5416 
5417 out:
5418 	mutex_unlock(&hr_qp->mutex);
5419 	return ret;
5420 }
5421 
5422 static inline int modify_qp_is_ok(struct hns_roce_qp *hr_qp)
5423 {
5424 	return ((hr_qp->ibqp.qp_type == IB_QPT_RC ||
5425 		 hr_qp->ibqp.qp_type == IB_QPT_UD ||
5426 		 hr_qp->ibqp.qp_type == IB_QPT_XRC_INI ||
5427 		 hr_qp->ibqp.qp_type == IB_QPT_XRC_TGT) &&
5428 		hr_qp->state != IB_QPS_RESET);
5429 }
5430 
5431 static int hns_roce_v2_destroy_qp_common(struct hns_roce_dev *hr_dev,
5432 					 struct hns_roce_qp *hr_qp,
5433 					 struct ib_udata *udata)
5434 {
5435 	struct ib_device *ibdev = &hr_dev->ib_dev;
5436 	struct hns_roce_cq *send_cq, *recv_cq;
5437 	unsigned long flags;
5438 	int ret = 0;
5439 
5440 	if (modify_qp_is_ok(hr_qp)) {
5441 		/* Modify qp to reset before destroying qp */
5442 		ret = hns_roce_v2_modify_qp(&hr_qp->ibqp, NULL, 0,
5443 					    hr_qp->state, IB_QPS_RESET, udata);
5444 		if (ret)
5445 			ibdev_err(ibdev,
5446 				  "failed to modify QP to RST, ret = %d.\n",
5447 				  ret);
5448 	}
5449 
5450 	send_cq = hr_qp->ibqp.send_cq ? to_hr_cq(hr_qp->ibqp.send_cq) : NULL;
5451 	recv_cq = hr_qp->ibqp.recv_cq ? to_hr_cq(hr_qp->ibqp.recv_cq) : NULL;
5452 
5453 	spin_lock_irqsave(&hr_dev->qp_list_lock, flags);
5454 	hns_roce_lock_cqs(send_cq, recv_cq);
5455 
5456 	if (!udata) {
5457 		if (recv_cq)
5458 			__hns_roce_v2_cq_clean(recv_cq, hr_qp->qpn,
5459 					       (hr_qp->ibqp.srq ?
5460 						to_hr_srq(hr_qp->ibqp.srq) :
5461 						NULL));
5462 
5463 		if (send_cq && send_cq != recv_cq)
5464 			__hns_roce_v2_cq_clean(send_cq, hr_qp->qpn, NULL);
5465 	}
5466 
5467 	hns_roce_qp_remove(hr_dev, hr_qp);
5468 
5469 	hns_roce_unlock_cqs(send_cq, recv_cq);
5470 	spin_unlock_irqrestore(&hr_dev->qp_list_lock, flags);
5471 
5472 	return ret;
5473 }
5474 
5475 int hns_roce_v2_destroy_qp(struct ib_qp *ibqp, struct ib_udata *udata)
5476 {
5477 	struct hns_roce_dev *hr_dev = to_hr_dev(ibqp->device);
5478 	struct hns_roce_qp *hr_qp = to_hr_qp(ibqp);
5479 	int ret;
5480 
5481 	ret = hns_roce_v2_destroy_qp_common(hr_dev, hr_qp, udata);
5482 	if (ret)
5483 		ibdev_err(&hr_dev->ib_dev,
5484 			  "failed to destroy QP, QPN = 0x%06lx, ret = %d.\n",
5485 			  hr_qp->qpn, ret);
5486 
5487 	hns_roce_qp_destroy(hr_dev, hr_qp, udata);
5488 
5489 	return 0;
5490 }
5491 
5492 static int hns_roce_v2_qp_flow_control_init(struct hns_roce_dev *hr_dev,
5493 					    struct hns_roce_qp *hr_qp)
5494 {
5495 	struct ib_device *ibdev = &hr_dev->ib_dev;
5496 	struct hns_roce_sccc_clr_done *resp;
5497 	struct hns_roce_sccc_clr *clr;
5498 	struct hns_roce_cmq_desc desc;
5499 	int ret, i;
5500 
5501 	if (hr_dev->pci_dev->revision >= PCI_REVISION_ID_HIP09)
5502 		return 0;
5503 
5504 	mutex_lock(&hr_dev->qp_table.scc_mutex);
5505 
5506 	/* set scc ctx clear done flag */
5507 	hns_roce_cmq_setup_basic_desc(&desc, HNS_ROCE_OPC_RESET_SCCC, false);
5508 	ret =  hns_roce_cmq_send(hr_dev, &desc, 1);
5509 	if (ret) {
5510 		ibdev_err(ibdev, "failed to reset SCC ctx, ret = %d.\n", ret);
5511 		goto out;
5512 	}
5513 
5514 	/* clear scc context */
5515 	hns_roce_cmq_setup_basic_desc(&desc, HNS_ROCE_OPC_CLR_SCCC, false);
5516 	clr = (struct hns_roce_sccc_clr *)desc.data;
5517 	clr->qpn = cpu_to_le32(hr_qp->qpn);
5518 	ret =  hns_roce_cmq_send(hr_dev, &desc, 1);
5519 	if (ret) {
5520 		ibdev_err(ibdev, "failed to clear SCC ctx, ret = %d.\n", ret);
5521 		goto out;
5522 	}
5523 
5524 	/* query scc context clear is done or not */
5525 	resp = (struct hns_roce_sccc_clr_done *)desc.data;
5526 	for (i = 0; i <= HNS_ROCE_CMQ_SCC_CLR_DONE_CNT; i++) {
5527 		hns_roce_cmq_setup_basic_desc(&desc,
5528 					      HNS_ROCE_OPC_QUERY_SCCC, true);
5529 		ret = hns_roce_cmq_send(hr_dev, &desc, 1);
5530 		if (ret) {
5531 			ibdev_err(ibdev, "failed to query clr cmq, ret = %d\n",
5532 				  ret);
5533 			goto out;
5534 		}
5535 
5536 		if (resp->clr_done)
5537 			goto out;
5538 
5539 		msleep(20);
5540 	}
5541 
5542 	ibdev_err(ibdev, "query SCC clr done flag overtime.\n");
5543 	ret = -ETIMEDOUT;
5544 
5545 out:
5546 	mutex_unlock(&hr_dev->qp_table.scc_mutex);
5547 	return ret;
5548 }
5549 
5550 #define DMA_IDX_SHIFT 3
5551 #define DMA_WQE_SHIFT 3
5552 
5553 static int hns_roce_v2_write_srqc_index_queue(struct hns_roce_srq *srq,
5554 					      struct hns_roce_srq_context *ctx)
5555 {
5556 	struct hns_roce_idx_que *idx_que = &srq->idx_que;
5557 	struct ib_device *ibdev = srq->ibsrq.device;
5558 	struct hns_roce_dev *hr_dev = to_hr_dev(ibdev);
5559 	u64 mtts_idx[MTT_MIN_COUNT] = {};
5560 	dma_addr_t dma_handle_idx = 0;
5561 	int ret;
5562 
5563 	/* Get physical address of idx que buf */
5564 	ret = hns_roce_mtr_find(hr_dev, &idx_que->mtr, 0, mtts_idx,
5565 				ARRAY_SIZE(mtts_idx), &dma_handle_idx);
5566 	if (ret < 1) {
5567 		ibdev_err(ibdev, "failed to find mtr for SRQ idx, ret = %d.\n",
5568 			  ret);
5569 		return -ENOBUFS;
5570 	}
5571 
5572 	hr_reg_write(ctx, SRQC_IDX_HOP_NUM,
5573 		     to_hr_hem_hopnum(hr_dev->caps.idx_hop_num, srq->wqe_cnt));
5574 
5575 	hr_reg_write(ctx, SRQC_IDX_BT_BA_L, dma_handle_idx >> DMA_IDX_SHIFT);
5576 	hr_reg_write(ctx, SRQC_IDX_BT_BA_H,
5577 		     upper_32_bits(dma_handle_idx >> DMA_IDX_SHIFT));
5578 
5579 	hr_reg_write(ctx, SRQC_IDX_BA_PG_SZ,
5580 		     to_hr_hw_page_shift(idx_que->mtr.hem_cfg.ba_pg_shift));
5581 	hr_reg_write(ctx, SRQC_IDX_BUF_PG_SZ,
5582 		     to_hr_hw_page_shift(idx_que->mtr.hem_cfg.buf_pg_shift));
5583 
5584 	hr_reg_write(ctx, SRQC_IDX_CUR_BLK_ADDR_L,
5585 		     to_hr_hw_page_addr(mtts_idx[0]));
5586 	hr_reg_write(ctx, SRQC_IDX_CUR_BLK_ADDR_H,
5587 		     upper_32_bits(to_hr_hw_page_addr(mtts_idx[0])));
5588 
5589 	hr_reg_write(ctx, SRQC_IDX_NXT_BLK_ADDR_L,
5590 		     to_hr_hw_page_addr(mtts_idx[1]));
5591 	hr_reg_write(ctx, SRQC_IDX_NXT_BLK_ADDR_H,
5592 		     upper_32_bits(to_hr_hw_page_addr(mtts_idx[1])));
5593 
5594 	return 0;
5595 }
5596 
5597 static int hns_roce_v2_write_srqc(struct hns_roce_srq *srq, void *mb_buf)
5598 {
5599 	struct ib_device *ibdev = srq->ibsrq.device;
5600 	struct hns_roce_dev *hr_dev = to_hr_dev(ibdev);
5601 	struct hns_roce_srq_context *ctx = mb_buf;
5602 	u64 mtts_wqe[MTT_MIN_COUNT] = {};
5603 	dma_addr_t dma_handle_wqe = 0;
5604 	int ret;
5605 
5606 	memset(ctx, 0, sizeof(*ctx));
5607 
5608 	/* Get the physical address of srq buf */
5609 	ret = hns_roce_mtr_find(hr_dev, &srq->buf_mtr, 0, mtts_wqe,
5610 				ARRAY_SIZE(mtts_wqe), &dma_handle_wqe);
5611 	if (ret < 1) {
5612 		ibdev_err(ibdev, "failed to find mtr for SRQ WQE, ret = %d.\n",
5613 			  ret);
5614 		return -ENOBUFS;
5615 	}
5616 
5617 	hr_reg_write(ctx, SRQC_SRQ_ST, 1);
5618 	hr_reg_write_bool(ctx, SRQC_SRQ_TYPE,
5619 			  srq->ibsrq.srq_type == IB_SRQT_XRC);
5620 	hr_reg_write(ctx, SRQC_PD, to_hr_pd(srq->ibsrq.pd)->pdn);
5621 	hr_reg_write(ctx, SRQC_SRQN, srq->srqn);
5622 	hr_reg_write(ctx, SRQC_XRCD, srq->xrcdn);
5623 	hr_reg_write(ctx, SRQC_XRC_CQN, srq->cqn);
5624 	hr_reg_write(ctx, SRQC_SHIFT, ilog2(srq->wqe_cnt));
5625 	hr_reg_write(ctx, SRQC_RQWS,
5626 		     srq->max_gs <= 0 ? 0 : fls(srq->max_gs - 1));
5627 
5628 	hr_reg_write(ctx, SRQC_WQE_HOP_NUM,
5629 		     to_hr_hem_hopnum(hr_dev->caps.srqwqe_hop_num,
5630 				      srq->wqe_cnt));
5631 
5632 	hr_reg_write(ctx, SRQC_WQE_BT_BA_L, dma_handle_wqe >> DMA_WQE_SHIFT);
5633 	hr_reg_write(ctx, SRQC_WQE_BT_BA_H,
5634 		     upper_32_bits(dma_handle_wqe >> DMA_WQE_SHIFT));
5635 
5636 	hr_reg_write(ctx, SRQC_WQE_BA_PG_SZ,
5637 		     to_hr_hw_page_shift(srq->buf_mtr.hem_cfg.ba_pg_shift));
5638 	hr_reg_write(ctx, SRQC_WQE_BUF_PG_SZ,
5639 		     to_hr_hw_page_shift(srq->buf_mtr.hem_cfg.buf_pg_shift));
5640 
5641 	return hns_roce_v2_write_srqc_index_queue(srq, ctx);
5642 }
5643 
5644 static int hns_roce_v2_modify_srq(struct ib_srq *ibsrq,
5645 				  struct ib_srq_attr *srq_attr,
5646 				  enum ib_srq_attr_mask srq_attr_mask,
5647 				  struct ib_udata *udata)
5648 {
5649 	struct hns_roce_dev *hr_dev = to_hr_dev(ibsrq->device);
5650 	struct hns_roce_srq *srq = to_hr_srq(ibsrq);
5651 	struct hns_roce_srq_context *srq_context;
5652 	struct hns_roce_srq_context *srqc_mask;
5653 	struct hns_roce_cmd_mailbox *mailbox;
5654 	int ret;
5655 
5656 	/* Resizing SRQs is not supported yet */
5657 	if (srq_attr_mask & IB_SRQ_MAX_WR)
5658 		return -EOPNOTSUPP;
5659 
5660 	if (srq_attr_mask & IB_SRQ_LIMIT) {
5661 		if (srq_attr->srq_limit > srq->wqe_cnt)
5662 			return -EINVAL;
5663 
5664 		mailbox = hns_roce_alloc_cmd_mailbox(hr_dev);
5665 		if (IS_ERR(mailbox))
5666 			return PTR_ERR(mailbox);
5667 
5668 		srq_context = mailbox->buf;
5669 		srqc_mask = (struct hns_roce_srq_context *)mailbox->buf + 1;
5670 
5671 		memset(srqc_mask, 0xff, sizeof(*srqc_mask));
5672 
5673 		hr_reg_write(srq_context, SRQC_LIMIT_WL, srq_attr->srq_limit);
5674 		hr_reg_clear(srqc_mask, SRQC_LIMIT_WL);
5675 
5676 		ret = hns_roce_cmd_mbox(hr_dev, mailbox->dma, 0,
5677 					HNS_ROCE_CMD_MODIFY_SRQC, srq->srqn);
5678 		hns_roce_free_cmd_mailbox(hr_dev, mailbox);
5679 		if (ret) {
5680 			ibdev_err(&hr_dev->ib_dev,
5681 				  "failed to handle cmd of modifying SRQ, ret = %d.\n",
5682 				  ret);
5683 			return ret;
5684 		}
5685 	}
5686 
5687 	return 0;
5688 }
5689 
5690 static int hns_roce_v2_query_srq(struct ib_srq *ibsrq, struct ib_srq_attr *attr)
5691 {
5692 	struct hns_roce_dev *hr_dev = to_hr_dev(ibsrq->device);
5693 	struct hns_roce_srq *srq = to_hr_srq(ibsrq);
5694 	struct hns_roce_srq_context *srq_context;
5695 	struct hns_roce_cmd_mailbox *mailbox;
5696 	int ret;
5697 
5698 	mailbox = hns_roce_alloc_cmd_mailbox(hr_dev);
5699 	if (IS_ERR(mailbox))
5700 		return PTR_ERR(mailbox);
5701 
5702 	srq_context = mailbox->buf;
5703 	ret = hns_roce_cmd_mbox(hr_dev, 0, mailbox->dma,
5704 				HNS_ROCE_CMD_QUERY_SRQC, srq->srqn);
5705 	if (ret) {
5706 		ibdev_err(&hr_dev->ib_dev,
5707 			  "failed to process cmd of querying SRQ, ret = %d.\n",
5708 			  ret);
5709 		goto out;
5710 	}
5711 
5712 	attr->srq_limit = hr_reg_read(srq_context, SRQC_LIMIT_WL);
5713 	attr->max_wr = srq->wqe_cnt;
5714 	attr->max_sge = srq->max_gs - srq->rsv_sge;
5715 
5716 out:
5717 	hns_roce_free_cmd_mailbox(hr_dev, mailbox);
5718 	return ret;
5719 }
5720 
5721 static int hns_roce_v2_modify_cq(struct ib_cq *cq, u16 cq_count, u16 cq_period)
5722 {
5723 	struct hns_roce_dev *hr_dev = to_hr_dev(cq->device);
5724 	struct hns_roce_v2_cq_context *cq_context;
5725 	struct hns_roce_cq *hr_cq = to_hr_cq(cq);
5726 	struct hns_roce_v2_cq_context *cqc_mask;
5727 	struct hns_roce_cmd_mailbox *mailbox;
5728 	int ret;
5729 
5730 	mailbox = hns_roce_alloc_cmd_mailbox(hr_dev);
5731 	if (IS_ERR(mailbox))
5732 		return PTR_ERR(mailbox);
5733 
5734 	cq_context = mailbox->buf;
5735 	cqc_mask = (struct hns_roce_v2_cq_context *)mailbox->buf + 1;
5736 
5737 	memset(cqc_mask, 0xff, sizeof(*cqc_mask));
5738 
5739 	hr_reg_write(cq_context, CQC_CQ_MAX_CNT, cq_count);
5740 	hr_reg_clear(cqc_mask, CQC_CQ_MAX_CNT);
5741 
5742 	if (hr_dev->pci_dev->revision == PCI_REVISION_ID_HIP08) {
5743 		if (cq_period * HNS_ROCE_CLOCK_ADJUST > USHRT_MAX) {
5744 			dev_info(hr_dev->dev,
5745 				 "cq_period(%u) reached the upper limit, adjusted to 65.\n",
5746 				 cq_period);
5747 			cq_period = HNS_ROCE_MAX_CQ_PERIOD;
5748 		}
5749 		cq_period *= HNS_ROCE_CLOCK_ADJUST;
5750 	}
5751 	hr_reg_write(cq_context, CQC_CQ_PERIOD, cq_period);
5752 	hr_reg_clear(cqc_mask, CQC_CQ_PERIOD);
5753 
5754 	ret = hns_roce_cmd_mbox(hr_dev, mailbox->dma, 0,
5755 				HNS_ROCE_CMD_MODIFY_CQC, hr_cq->cqn);
5756 	hns_roce_free_cmd_mailbox(hr_dev, mailbox);
5757 	if (ret)
5758 		ibdev_err(&hr_dev->ib_dev,
5759 			  "failed to process cmd when modifying CQ, ret = %d.\n",
5760 			  ret);
5761 
5762 	return ret;
5763 }
5764 
5765 static int hns_roce_v2_query_cqc(struct hns_roce_dev *hr_dev, u32 cqn,
5766 				 void *buffer)
5767 {
5768 	struct hns_roce_v2_cq_context *context;
5769 	struct hns_roce_cmd_mailbox *mailbox;
5770 	int ret;
5771 
5772 	mailbox = hns_roce_alloc_cmd_mailbox(hr_dev);
5773 	if (IS_ERR(mailbox))
5774 		return PTR_ERR(mailbox);
5775 
5776 	context = mailbox->buf;
5777 	ret = hns_roce_cmd_mbox(hr_dev, 0, mailbox->dma,
5778 				HNS_ROCE_CMD_QUERY_CQC, cqn);
5779 	if (ret) {
5780 		ibdev_err(&hr_dev->ib_dev,
5781 			  "failed to process cmd when querying CQ, ret = %d.\n",
5782 			  ret);
5783 		goto err_mailbox;
5784 	}
5785 
5786 	memcpy(buffer, context, sizeof(*context));
5787 
5788 err_mailbox:
5789 	hns_roce_free_cmd_mailbox(hr_dev, mailbox);
5790 
5791 	return ret;
5792 }
5793 
5794 static int hns_roce_v2_query_mpt(struct hns_roce_dev *hr_dev, u32 key,
5795 				 void *buffer)
5796 {
5797 	struct hns_roce_v2_mpt_entry *context;
5798 	struct hns_roce_cmd_mailbox *mailbox;
5799 	int ret;
5800 
5801 	mailbox = hns_roce_alloc_cmd_mailbox(hr_dev);
5802 	if (IS_ERR(mailbox))
5803 		return PTR_ERR(mailbox);
5804 
5805 	context = mailbox->buf;
5806 	ret = hns_roce_cmd_mbox(hr_dev, 0, mailbox->dma, HNS_ROCE_CMD_QUERY_MPT,
5807 				key_to_hw_index(key));
5808 	if (ret) {
5809 		ibdev_err(&hr_dev->ib_dev,
5810 			  "failed to process cmd when querying MPT, ret = %d.\n",
5811 			  ret);
5812 		goto err_mailbox;
5813 	}
5814 
5815 	memcpy(buffer, context, sizeof(*context));
5816 
5817 err_mailbox:
5818 	hns_roce_free_cmd_mailbox(hr_dev, mailbox);
5819 
5820 	return ret;
5821 }
5822 
5823 static void hns_roce_irq_work_handle(struct work_struct *work)
5824 {
5825 	struct hns_roce_work *irq_work =
5826 				container_of(work, struct hns_roce_work, work);
5827 	struct ib_device *ibdev = &irq_work->hr_dev->ib_dev;
5828 
5829 	switch (irq_work->event_type) {
5830 	case HNS_ROCE_EVENT_TYPE_PATH_MIG:
5831 		ibdev_info(ibdev, "path migrated succeeded.\n");
5832 		break;
5833 	case HNS_ROCE_EVENT_TYPE_PATH_MIG_FAILED:
5834 		ibdev_warn(ibdev, "path migration failed.\n");
5835 		break;
5836 	case HNS_ROCE_EVENT_TYPE_COMM_EST:
5837 		break;
5838 	case HNS_ROCE_EVENT_TYPE_SQ_DRAINED:
5839 		ibdev_dbg(ibdev, "send queue drained.\n");
5840 		break;
5841 	case HNS_ROCE_EVENT_TYPE_WQ_CATAS_ERROR:
5842 		ibdev_err(ibdev, "local work queue 0x%x catast error, sub_event type is: %d\n",
5843 			  irq_work->queue_num, irq_work->sub_type);
5844 		break;
5845 	case HNS_ROCE_EVENT_TYPE_INV_REQ_LOCAL_WQ_ERROR:
5846 		ibdev_err(ibdev, "invalid request local work queue 0x%x error.\n",
5847 			  irq_work->queue_num);
5848 		break;
5849 	case HNS_ROCE_EVENT_TYPE_LOCAL_WQ_ACCESS_ERROR:
5850 		ibdev_err(ibdev, "local access violation work queue 0x%x error, sub_event type is: %d\n",
5851 			  irq_work->queue_num, irq_work->sub_type);
5852 		break;
5853 	case HNS_ROCE_EVENT_TYPE_SRQ_LIMIT_REACH:
5854 		ibdev_dbg(ibdev, "SRQ limit reach.\n");
5855 		break;
5856 	case HNS_ROCE_EVENT_TYPE_SRQ_LAST_WQE_REACH:
5857 		ibdev_dbg(ibdev, "SRQ last wqe reach.\n");
5858 		break;
5859 	case HNS_ROCE_EVENT_TYPE_SRQ_CATAS_ERROR:
5860 		ibdev_err(ibdev, "SRQ catas error.\n");
5861 		break;
5862 	case HNS_ROCE_EVENT_TYPE_CQ_ACCESS_ERROR:
5863 		ibdev_err(ibdev, "CQ 0x%x access err.\n", irq_work->queue_num);
5864 		break;
5865 	case HNS_ROCE_EVENT_TYPE_CQ_OVERFLOW:
5866 		ibdev_warn(ibdev, "CQ 0x%x overflow\n", irq_work->queue_num);
5867 		break;
5868 	case HNS_ROCE_EVENT_TYPE_DB_OVERFLOW:
5869 		ibdev_warn(ibdev, "DB overflow.\n");
5870 		break;
5871 	case HNS_ROCE_EVENT_TYPE_FLR:
5872 		ibdev_warn(ibdev, "function level reset.\n");
5873 		break;
5874 	case HNS_ROCE_EVENT_TYPE_XRCD_VIOLATION:
5875 		ibdev_err(ibdev, "xrc domain violation error.\n");
5876 		break;
5877 	case HNS_ROCE_EVENT_TYPE_INVALID_XRCETH:
5878 		ibdev_err(ibdev, "invalid xrceth error.\n");
5879 		break;
5880 	default:
5881 		break;
5882 	}
5883 
5884 	kfree(irq_work);
5885 }
5886 
5887 static void hns_roce_v2_init_irq_work(struct hns_roce_dev *hr_dev,
5888 				      struct hns_roce_eq *eq, u32 queue_num)
5889 {
5890 	struct hns_roce_work *irq_work;
5891 
5892 	irq_work = kzalloc(sizeof(struct hns_roce_work), GFP_ATOMIC);
5893 	if (!irq_work)
5894 		return;
5895 
5896 	INIT_WORK(&irq_work->work, hns_roce_irq_work_handle);
5897 	irq_work->hr_dev = hr_dev;
5898 	irq_work->event_type = eq->event_type;
5899 	irq_work->sub_type = eq->sub_type;
5900 	irq_work->queue_num = queue_num;
5901 	queue_work(hr_dev->irq_workq, &irq_work->work);
5902 }
5903 
5904 static void update_eq_db(struct hns_roce_eq *eq)
5905 {
5906 	struct hns_roce_dev *hr_dev = eq->hr_dev;
5907 	struct hns_roce_v2_db eq_db = {};
5908 
5909 	if (eq->type_flag == HNS_ROCE_AEQ) {
5910 		hr_reg_write(&eq_db, EQ_DB_CMD,
5911 			     eq->arm_st == HNS_ROCE_V2_EQ_ALWAYS_ARMED ?
5912 			     HNS_ROCE_EQ_DB_CMD_AEQ :
5913 			     HNS_ROCE_EQ_DB_CMD_AEQ_ARMED);
5914 	} else {
5915 		hr_reg_write(&eq_db, EQ_DB_TAG, eq->eqn);
5916 
5917 		hr_reg_write(&eq_db, EQ_DB_CMD,
5918 			     eq->arm_st == HNS_ROCE_V2_EQ_ALWAYS_ARMED ?
5919 			     HNS_ROCE_EQ_DB_CMD_CEQ :
5920 			     HNS_ROCE_EQ_DB_CMD_CEQ_ARMED);
5921 	}
5922 
5923 	hr_reg_write(&eq_db, EQ_DB_CI, eq->cons_index);
5924 
5925 	hns_roce_write64(hr_dev, (__le32 *)&eq_db, eq->db_reg);
5926 }
5927 
5928 static struct hns_roce_aeqe *next_aeqe_sw_v2(struct hns_roce_eq *eq)
5929 {
5930 	struct hns_roce_aeqe *aeqe;
5931 
5932 	aeqe = hns_roce_buf_offset(eq->mtr.kmem,
5933 				   (eq->cons_index & (eq->entries - 1)) *
5934 				   eq->eqe_size);
5935 
5936 	return (hr_reg_read(aeqe, AEQE_OWNER) ^
5937 		!!(eq->cons_index & eq->entries)) ? aeqe : NULL;
5938 }
5939 
5940 static irqreturn_t hns_roce_v2_aeq_int(struct hns_roce_dev *hr_dev,
5941 				       struct hns_roce_eq *eq)
5942 {
5943 	struct device *dev = hr_dev->dev;
5944 	struct hns_roce_aeqe *aeqe = next_aeqe_sw_v2(eq);
5945 	irqreturn_t aeqe_found = IRQ_NONE;
5946 	int event_type;
5947 	u32 queue_num;
5948 	int sub_type;
5949 
5950 	while (aeqe) {
5951 		/* Make sure we read AEQ entry after we have checked the
5952 		 * ownership bit
5953 		 */
5954 		dma_rmb();
5955 
5956 		event_type = hr_reg_read(aeqe, AEQE_EVENT_TYPE);
5957 		sub_type = hr_reg_read(aeqe, AEQE_SUB_TYPE);
5958 		queue_num = hr_reg_read(aeqe, AEQE_EVENT_QUEUE_NUM);
5959 
5960 		switch (event_type) {
5961 		case HNS_ROCE_EVENT_TYPE_PATH_MIG:
5962 		case HNS_ROCE_EVENT_TYPE_PATH_MIG_FAILED:
5963 		case HNS_ROCE_EVENT_TYPE_COMM_EST:
5964 		case HNS_ROCE_EVENT_TYPE_SQ_DRAINED:
5965 		case HNS_ROCE_EVENT_TYPE_WQ_CATAS_ERROR:
5966 		case HNS_ROCE_EVENT_TYPE_SRQ_LAST_WQE_REACH:
5967 		case HNS_ROCE_EVENT_TYPE_INV_REQ_LOCAL_WQ_ERROR:
5968 		case HNS_ROCE_EVENT_TYPE_LOCAL_WQ_ACCESS_ERROR:
5969 		case HNS_ROCE_EVENT_TYPE_XRCD_VIOLATION:
5970 		case HNS_ROCE_EVENT_TYPE_INVALID_XRCETH:
5971 			hns_roce_qp_event(hr_dev, queue_num, event_type);
5972 			break;
5973 		case HNS_ROCE_EVENT_TYPE_SRQ_LIMIT_REACH:
5974 		case HNS_ROCE_EVENT_TYPE_SRQ_CATAS_ERROR:
5975 			hns_roce_srq_event(hr_dev, queue_num, event_type);
5976 			break;
5977 		case HNS_ROCE_EVENT_TYPE_CQ_ACCESS_ERROR:
5978 		case HNS_ROCE_EVENT_TYPE_CQ_OVERFLOW:
5979 			hns_roce_cq_event(hr_dev, queue_num, event_type);
5980 			break;
5981 		case HNS_ROCE_EVENT_TYPE_MB:
5982 			hns_roce_cmd_event(hr_dev,
5983 					le16_to_cpu(aeqe->event.cmd.token),
5984 					aeqe->event.cmd.status,
5985 					le64_to_cpu(aeqe->event.cmd.out_param));
5986 			break;
5987 		case HNS_ROCE_EVENT_TYPE_DB_OVERFLOW:
5988 		case HNS_ROCE_EVENT_TYPE_FLR:
5989 			break;
5990 		default:
5991 			dev_err(dev, "unhandled event %d on EQ %d at idx %u.\n",
5992 				event_type, eq->eqn, eq->cons_index);
5993 			break;
5994 		}
5995 
5996 		eq->event_type = event_type;
5997 		eq->sub_type = sub_type;
5998 		++eq->cons_index;
5999 		aeqe_found = IRQ_HANDLED;
6000 
6001 		hns_roce_v2_init_irq_work(hr_dev, eq, queue_num);
6002 
6003 		aeqe = next_aeqe_sw_v2(eq);
6004 	}
6005 
6006 	update_eq_db(eq);
6007 
6008 	return IRQ_RETVAL(aeqe_found);
6009 }
6010 
6011 static struct hns_roce_ceqe *next_ceqe_sw_v2(struct hns_roce_eq *eq)
6012 {
6013 	struct hns_roce_ceqe *ceqe;
6014 
6015 	ceqe = hns_roce_buf_offset(eq->mtr.kmem,
6016 				   (eq->cons_index & (eq->entries - 1)) *
6017 				   eq->eqe_size);
6018 
6019 	return (hr_reg_read(ceqe, CEQE_OWNER) ^
6020 		!!(eq->cons_index & eq->entries)) ? ceqe : NULL;
6021 }
6022 
6023 static irqreturn_t hns_roce_v2_ceq_int(struct hns_roce_dev *hr_dev,
6024 				       struct hns_roce_eq *eq)
6025 {
6026 	struct hns_roce_ceqe *ceqe = next_ceqe_sw_v2(eq);
6027 	irqreturn_t ceqe_found = IRQ_NONE;
6028 	u32 cqn;
6029 
6030 	while (ceqe) {
6031 		/* Make sure we read CEQ entry after we have checked the
6032 		 * ownership bit
6033 		 */
6034 		dma_rmb();
6035 
6036 		cqn = hr_reg_read(ceqe, CEQE_CQN);
6037 
6038 		hns_roce_cq_completion(hr_dev, cqn);
6039 
6040 		++eq->cons_index;
6041 		ceqe_found = IRQ_HANDLED;
6042 
6043 		ceqe = next_ceqe_sw_v2(eq);
6044 	}
6045 
6046 	update_eq_db(eq);
6047 
6048 	return IRQ_RETVAL(ceqe_found);
6049 }
6050 
6051 static irqreturn_t hns_roce_v2_msix_interrupt_eq(int irq, void *eq_ptr)
6052 {
6053 	struct hns_roce_eq *eq = eq_ptr;
6054 	struct hns_roce_dev *hr_dev = eq->hr_dev;
6055 	irqreturn_t int_work;
6056 
6057 	if (eq->type_flag == HNS_ROCE_CEQ)
6058 		/* Completion event interrupt */
6059 		int_work = hns_roce_v2_ceq_int(hr_dev, eq);
6060 	else
6061 		/* Asynchronous event interrupt */
6062 		int_work = hns_roce_v2_aeq_int(hr_dev, eq);
6063 
6064 	return IRQ_RETVAL(int_work);
6065 }
6066 
6067 static irqreturn_t abnormal_interrupt_basic(struct hns_roce_dev *hr_dev,
6068 					    u32 int_st)
6069 {
6070 	struct pci_dev *pdev = hr_dev->pci_dev;
6071 	struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
6072 	const struct hnae3_ae_ops *ops = ae_dev->ops;
6073 	enum hnae3_reset_type reset_type;
6074 	irqreturn_t int_work = IRQ_NONE;
6075 	u32 int_en;
6076 
6077 	int_en = roce_read(hr_dev, ROCEE_VF_ABN_INT_EN_REG);
6078 
6079 	if (int_st & BIT(HNS_ROCE_V2_VF_INT_ST_AEQ_OVERFLOW_S)) {
6080 		dev_err(hr_dev->dev, "AEQ overflow!\n");
6081 
6082 		roce_write(hr_dev, ROCEE_VF_ABN_INT_ST_REG,
6083 			   1 << HNS_ROCE_V2_VF_INT_ST_AEQ_OVERFLOW_S);
6084 
6085 		reset_type = hr_dev->is_vf ?
6086 			     HNAE3_VF_FUNC_RESET : HNAE3_FUNC_RESET;
6087 
6088 		/* Set reset level for reset_event() */
6089 		if (ops->set_default_reset_request)
6090 			ops->set_default_reset_request(ae_dev, reset_type);
6091 		if (ops->reset_event)
6092 			ops->reset_event(pdev, NULL);
6093 
6094 		int_en |= 1 << HNS_ROCE_V2_VF_ABN_INT_EN_S;
6095 		roce_write(hr_dev, ROCEE_VF_ABN_INT_EN_REG, int_en);
6096 
6097 		int_work = IRQ_HANDLED;
6098 	} else {
6099 		dev_err(hr_dev->dev, "there is no basic abn irq found.\n");
6100 	}
6101 
6102 	return IRQ_RETVAL(int_work);
6103 }
6104 
6105 static int fmea_ram_ecc_query(struct hns_roce_dev *hr_dev,
6106 			       struct fmea_ram_ecc *ecc_info)
6107 {
6108 	struct hns_roce_cmq_desc desc;
6109 	struct hns_roce_cmq_req *req = (struct hns_roce_cmq_req *)desc.data;
6110 	int ret;
6111 
6112 	hns_roce_cmq_setup_basic_desc(&desc, HNS_ROCE_QUERY_RAM_ECC, true);
6113 	ret = hns_roce_cmq_send(hr_dev, &desc, 1);
6114 	if (ret)
6115 		return ret;
6116 
6117 	ecc_info->is_ecc_err = hr_reg_read(req, QUERY_RAM_ECC_1BIT_ERR);
6118 	ecc_info->res_type = hr_reg_read(req, QUERY_RAM_ECC_RES_TYPE);
6119 	ecc_info->index = hr_reg_read(req, QUERY_RAM_ECC_TAG);
6120 
6121 	return 0;
6122 }
6123 
6124 static int fmea_recover_gmv(struct hns_roce_dev *hr_dev, u32 idx)
6125 {
6126 	struct hns_roce_cmq_desc desc;
6127 	struct hns_roce_cmq_req *req = (struct hns_roce_cmq_req *)desc.data;
6128 	u32 addr_upper;
6129 	u32 addr_low;
6130 	int ret;
6131 
6132 	hns_roce_cmq_setup_basic_desc(&desc, HNS_ROCE_OPC_CFG_GMV_BT, true);
6133 	hr_reg_write(req, CFG_GMV_BT_IDX, idx);
6134 
6135 	ret = hns_roce_cmq_send(hr_dev, &desc, 1);
6136 	if (ret) {
6137 		dev_err(hr_dev->dev,
6138 			"failed to execute cmd to read gmv, ret = %d.\n", ret);
6139 		return ret;
6140 	}
6141 
6142 	addr_low =  hr_reg_read(req, CFG_GMV_BT_BA_L);
6143 	addr_upper = hr_reg_read(req, CFG_GMV_BT_BA_H);
6144 
6145 	hns_roce_cmq_setup_basic_desc(&desc, HNS_ROCE_OPC_CFG_GMV_BT, false);
6146 	hr_reg_write(req, CFG_GMV_BT_BA_L, addr_low);
6147 	hr_reg_write(req, CFG_GMV_BT_BA_H, addr_upper);
6148 	hr_reg_write(req, CFG_GMV_BT_IDX, idx);
6149 
6150 	return hns_roce_cmq_send(hr_dev, &desc, 1);
6151 }
6152 
6153 static u64 fmea_get_ram_res_addr(u32 res_type, __le64 *data)
6154 {
6155 	if (res_type == ECC_RESOURCE_QPC_TIMER ||
6156 	    res_type == ECC_RESOURCE_CQC_TIMER ||
6157 	    res_type == ECC_RESOURCE_SCCC)
6158 		return le64_to_cpu(*data);
6159 
6160 	return le64_to_cpu(*data) << HNS_HW_PAGE_SHIFT;
6161 }
6162 
6163 static int fmea_recover_others(struct hns_roce_dev *hr_dev, u32 res_type,
6164 			       u32 index)
6165 {
6166 	u8 write_bt0_op = fmea_ram_res[res_type].write_bt0_op;
6167 	u8 read_bt0_op = fmea_ram_res[res_type].read_bt0_op;
6168 	struct hns_roce_cmd_mailbox *mailbox;
6169 	u64 addr;
6170 	int ret;
6171 
6172 	mailbox = hns_roce_alloc_cmd_mailbox(hr_dev);
6173 	if (IS_ERR(mailbox))
6174 		return PTR_ERR(mailbox);
6175 
6176 	ret = hns_roce_cmd_mbox(hr_dev, 0, mailbox->dma, read_bt0_op, index);
6177 	if (ret) {
6178 		dev_err(hr_dev->dev,
6179 			"failed to execute cmd to read fmea ram, ret = %d.\n",
6180 			ret);
6181 		goto out;
6182 	}
6183 
6184 	addr = fmea_get_ram_res_addr(res_type, mailbox->buf);
6185 
6186 	ret = hns_roce_cmd_mbox(hr_dev, addr, 0, write_bt0_op, index);
6187 	if (ret)
6188 		dev_err(hr_dev->dev,
6189 			"failed to execute cmd to write fmea ram, ret = %d.\n",
6190 			ret);
6191 
6192 out:
6193 	hns_roce_free_cmd_mailbox(hr_dev, mailbox);
6194 	return ret;
6195 }
6196 
6197 static void fmea_ram_ecc_recover(struct hns_roce_dev *hr_dev,
6198 				 struct fmea_ram_ecc *ecc_info)
6199 {
6200 	u32 res_type = ecc_info->res_type;
6201 	u32 index = ecc_info->index;
6202 	int ret;
6203 
6204 	BUILD_BUG_ON(ARRAY_SIZE(fmea_ram_res) != ECC_RESOURCE_COUNT);
6205 
6206 	if (res_type >= ECC_RESOURCE_COUNT) {
6207 		dev_err(hr_dev->dev, "unsupported fmea ram ecc type %u.\n",
6208 			res_type);
6209 		return;
6210 	}
6211 
6212 	if (res_type == ECC_RESOURCE_GMV)
6213 		ret = fmea_recover_gmv(hr_dev, index);
6214 	else
6215 		ret = fmea_recover_others(hr_dev, res_type, index);
6216 	if (ret)
6217 		dev_err(hr_dev->dev,
6218 			"failed to recover %s, index = %u, ret = %d.\n",
6219 			fmea_ram_res[res_type].name, index, ret);
6220 }
6221 
6222 static void fmea_ram_ecc_work(struct work_struct *ecc_work)
6223 {
6224 	struct hns_roce_dev *hr_dev =
6225 		container_of(ecc_work, struct hns_roce_dev, ecc_work);
6226 	struct fmea_ram_ecc ecc_info = {};
6227 
6228 	if (fmea_ram_ecc_query(hr_dev, &ecc_info)) {
6229 		dev_err(hr_dev->dev, "failed to query fmea ram ecc.\n");
6230 		return;
6231 	}
6232 
6233 	if (!ecc_info.is_ecc_err) {
6234 		dev_err(hr_dev->dev, "there is no fmea ram ecc err found.\n");
6235 		return;
6236 	}
6237 
6238 	fmea_ram_ecc_recover(hr_dev, &ecc_info);
6239 }
6240 
6241 static irqreturn_t hns_roce_v2_msix_interrupt_abn(int irq, void *dev_id)
6242 {
6243 	struct hns_roce_dev *hr_dev = dev_id;
6244 	irqreturn_t int_work = IRQ_NONE;
6245 	u32 int_st;
6246 
6247 	int_st = roce_read(hr_dev, ROCEE_VF_ABN_INT_ST_REG);
6248 
6249 	if (int_st) {
6250 		int_work = abnormal_interrupt_basic(hr_dev, int_st);
6251 	} else if (hr_dev->pci_dev->revision >= PCI_REVISION_ID_HIP09) {
6252 		queue_work(hr_dev->irq_workq, &hr_dev->ecc_work);
6253 		int_work = IRQ_HANDLED;
6254 	} else {
6255 		dev_err(hr_dev->dev, "there is no abnormal irq found.\n");
6256 	}
6257 
6258 	return IRQ_RETVAL(int_work);
6259 }
6260 
6261 static void hns_roce_v2_int_mask_enable(struct hns_roce_dev *hr_dev,
6262 					int eq_num, u32 enable_flag)
6263 {
6264 	int i;
6265 
6266 	for (i = 0; i < eq_num; i++)
6267 		roce_write(hr_dev, ROCEE_VF_EVENT_INT_EN_REG +
6268 			   i * EQ_REG_OFFSET, enable_flag);
6269 
6270 	roce_write(hr_dev, ROCEE_VF_ABN_INT_EN_REG, enable_flag);
6271 	roce_write(hr_dev, ROCEE_VF_ABN_INT_CFG_REG, enable_flag);
6272 }
6273 
6274 static void free_eq_buf(struct hns_roce_dev *hr_dev, struct hns_roce_eq *eq)
6275 {
6276 	hns_roce_mtr_destroy(hr_dev, &eq->mtr);
6277 }
6278 
6279 static void hns_roce_v2_destroy_eqc(struct hns_roce_dev *hr_dev,
6280 				    struct hns_roce_eq *eq)
6281 {
6282 	struct device *dev = hr_dev->dev;
6283 	int eqn = eq->eqn;
6284 	int ret;
6285 	u8 cmd;
6286 
6287 	if (eqn < hr_dev->caps.num_comp_vectors)
6288 		cmd = HNS_ROCE_CMD_DESTROY_CEQC;
6289 	else
6290 		cmd = HNS_ROCE_CMD_DESTROY_AEQC;
6291 
6292 	ret = hns_roce_destroy_hw_ctx(hr_dev, cmd, eqn & HNS_ROCE_V2_EQN_M);
6293 	if (ret)
6294 		dev_err(dev, "[mailbox cmd] destroy eqc(%d) failed.\n", eqn);
6295 
6296 	free_eq_buf(hr_dev, eq);
6297 }
6298 
6299 static void init_eq_config(struct hns_roce_dev *hr_dev, struct hns_roce_eq *eq)
6300 {
6301 	eq->db_reg = hr_dev->reg_base + ROCEE_VF_EQ_DB_CFG0_REG;
6302 	eq->cons_index = 0;
6303 	eq->over_ignore = HNS_ROCE_V2_EQ_OVER_IGNORE_0;
6304 	eq->coalesce = HNS_ROCE_V2_EQ_COALESCE_0;
6305 	eq->arm_st = HNS_ROCE_V2_EQ_ALWAYS_ARMED;
6306 	eq->shift = ilog2((unsigned int)eq->entries);
6307 }
6308 
6309 static int config_eqc(struct hns_roce_dev *hr_dev, struct hns_roce_eq *eq,
6310 		      void *mb_buf)
6311 {
6312 	u64 eqe_ba[MTT_MIN_COUNT] = { 0 };
6313 	struct hns_roce_eq_context *eqc;
6314 	u64 bt_ba = 0;
6315 	int count;
6316 
6317 	eqc = mb_buf;
6318 	memset(eqc, 0, sizeof(struct hns_roce_eq_context));
6319 
6320 	init_eq_config(hr_dev, eq);
6321 
6322 	/* if not multi-hop, eqe buffer only use one trunk */
6323 	count = hns_roce_mtr_find(hr_dev, &eq->mtr, 0, eqe_ba, MTT_MIN_COUNT,
6324 				  &bt_ba);
6325 	if (count < 1) {
6326 		dev_err(hr_dev->dev, "failed to find EQE mtr\n");
6327 		return -ENOBUFS;
6328 	}
6329 
6330 	hr_reg_write(eqc, EQC_EQ_ST, HNS_ROCE_V2_EQ_STATE_VALID);
6331 	hr_reg_write(eqc, EQC_EQE_HOP_NUM, eq->hop_num);
6332 	hr_reg_write(eqc, EQC_OVER_IGNORE, eq->over_ignore);
6333 	hr_reg_write(eqc, EQC_COALESCE, eq->coalesce);
6334 	hr_reg_write(eqc, EQC_ARM_ST, eq->arm_st);
6335 	hr_reg_write(eqc, EQC_EQN, eq->eqn);
6336 	hr_reg_write(eqc, EQC_EQE_CNT, HNS_ROCE_EQ_INIT_EQE_CNT);
6337 	hr_reg_write(eqc, EQC_EQE_BA_PG_SZ,
6338 		     to_hr_hw_page_shift(eq->mtr.hem_cfg.ba_pg_shift));
6339 	hr_reg_write(eqc, EQC_EQE_BUF_PG_SZ,
6340 		     to_hr_hw_page_shift(eq->mtr.hem_cfg.buf_pg_shift));
6341 	hr_reg_write(eqc, EQC_EQ_PROD_INDX, HNS_ROCE_EQ_INIT_PROD_IDX);
6342 	hr_reg_write(eqc, EQC_EQ_MAX_CNT, eq->eq_max_cnt);
6343 
6344 	if (hr_dev->pci_dev->revision == PCI_REVISION_ID_HIP08) {
6345 		if (eq->eq_period * HNS_ROCE_CLOCK_ADJUST > USHRT_MAX) {
6346 			dev_info(hr_dev->dev, "eq_period(%u) reached the upper limit, adjusted to 65.\n",
6347 				 eq->eq_period);
6348 			eq->eq_period = HNS_ROCE_MAX_EQ_PERIOD;
6349 		}
6350 		eq->eq_period *= HNS_ROCE_CLOCK_ADJUST;
6351 	}
6352 
6353 	hr_reg_write(eqc, EQC_EQ_PERIOD, eq->eq_period);
6354 	hr_reg_write(eqc, EQC_EQE_REPORT_TIMER, HNS_ROCE_EQ_INIT_REPORT_TIMER);
6355 	hr_reg_write(eqc, EQC_EQE_BA_L, bt_ba >> 3);
6356 	hr_reg_write(eqc, EQC_EQE_BA_H, bt_ba >> 35);
6357 	hr_reg_write(eqc, EQC_SHIFT, eq->shift);
6358 	hr_reg_write(eqc, EQC_MSI_INDX, HNS_ROCE_EQ_INIT_MSI_IDX);
6359 	hr_reg_write(eqc, EQC_CUR_EQE_BA_L, eqe_ba[0] >> 12);
6360 	hr_reg_write(eqc, EQC_CUR_EQE_BA_M, eqe_ba[0] >> 28);
6361 	hr_reg_write(eqc, EQC_CUR_EQE_BA_H, eqe_ba[0] >> 60);
6362 	hr_reg_write(eqc, EQC_EQ_CONS_INDX, HNS_ROCE_EQ_INIT_CONS_IDX);
6363 	hr_reg_write(eqc, EQC_NEX_EQE_BA_L, eqe_ba[1] >> 12);
6364 	hr_reg_write(eqc, EQC_NEX_EQE_BA_H, eqe_ba[1] >> 44);
6365 	hr_reg_write(eqc, EQC_EQE_SIZE, eq->eqe_size == HNS_ROCE_V3_EQE_SIZE);
6366 
6367 	return 0;
6368 }
6369 
6370 static int alloc_eq_buf(struct hns_roce_dev *hr_dev, struct hns_roce_eq *eq)
6371 {
6372 	struct hns_roce_buf_attr buf_attr = {};
6373 	int err;
6374 
6375 	if (hr_dev->caps.eqe_hop_num == HNS_ROCE_HOP_NUM_0)
6376 		eq->hop_num = 0;
6377 	else
6378 		eq->hop_num = hr_dev->caps.eqe_hop_num;
6379 
6380 	buf_attr.page_shift = hr_dev->caps.eqe_buf_pg_sz + PAGE_SHIFT;
6381 	buf_attr.region[0].size = eq->entries * eq->eqe_size;
6382 	buf_attr.region[0].hopnum = eq->hop_num;
6383 	buf_attr.region_count = 1;
6384 
6385 	err = hns_roce_mtr_create(hr_dev, &eq->mtr, &buf_attr,
6386 				  hr_dev->caps.eqe_ba_pg_sz + PAGE_SHIFT, NULL,
6387 				  0);
6388 	if (err)
6389 		dev_err(hr_dev->dev, "failed to alloc EQE mtr, err %d\n", err);
6390 
6391 	return err;
6392 }
6393 
6394 static int hns_roce_v2_create_eq(struct hns_roce_dev *hr_dev,
6395 				 struct hns_roce_eq *eq, u8 eq_cmd)
6396 {
6397 	struct hns_roce_cmd_mailbox *mailbox;
6398 	int ret;
6399 
6400 	/* Allocate mailbox memory */
6401 	mailbox = hns_roce_alloc_cmd_mailbox(hr_dev);
6402 	if (IS_ERR(mailbox))
6403 		return PTR_ERR(mailbox);
6404 
6405 	ret = alloc_eq_buf(hr_dev, eq);
6406 	if (ret)
6407 		goto free_cmd_mbox;
6408 
6409 	ret = config_eqc(hr_dev, eq, mailbox->buf);
6410 	if (ret)
6411 		goto err_cmd_mbox;
6412 
6413 	ret = hns_roce_create_hw_ctx(hr_dev, mailbox, eq_cmd, eq->eqn);
6414 	if (ret) {
6415 		dev_err(hr_dev->dev, "[mailbox cmd] create eqc failed.\n");
6416 		goto err_cmd_mbox;
6417 	}
6418 
6419 	hns_roce_free_cmd_mailbox(hr_dev, mailbox);
6420 
6421 	return 0;
6422 
6423 err_cmd_mbox:
6424 	free_eq_buf(hr_dev, eq);
6425 
6426 free_cmd_mbox:
6427 	hns_roce_free_cmd_mailbox(hr_dev, mailbox);
6428 
6429 	return ret;
6430 }
6431 
6432 static int __hns_roce_request_irq(struct hns_roce_dev *hr_dev, int irq_num,
6433 				  int comp_num, int aeq_num, int other_num)
6434 {
6435 	struct hns_roce_eq_table *eq_table = &hr_dev->eq_table;
6436 	int i, j;
6437 	int ret;
6438 
6439 	for (i = 0; i < irq_num; i++) {
6440 		hr_dev->irq_names[i] = kzalloc(HNS_ROCE_INT_NAME_LEN,
6441 					       GFP_KERNEL);
6442 		if (!hr_dev->irq_names[i]) {
6443 			ret = -ENOMEM;
6444 			goto err_kzalloc_failed;
6445 		}
6446 	}
6447 
6448 	/* irq contains: abnormal + AEQ + CEQ */
6449 	for (j = 0; j < other_num; j++)
6450 		snprintf((char *)hr_dev->irq_names[j], HNS_ROCE_INT_NAME_LEN,
6451 			 "hns-abn-%d", j);
6452 
6453 	for (j = other_num; j < (other_num + aeq_num); j++)
6454 		snprintf((char *)hr_dev->irq_names[j], HNS_ROCE_INT_NAME_LEN,
6455 			 "hns-aeq-%d", j - other_num);
6456 
6457 	for (j = (other_num + aeq_num); j < irq_num; j++)
6458 		snprintf((char *)hr_dev->irq_names[j], HNS_ROCE_INT_NAME_LEN,
6459 			 "hns-ceq-%d", j - other_num - aeq_num);
6460 
6461 	for (j = 0; j < irq_num; j++) {
6462 		if (j < other_num)
6463 			ret = request_irq(hr_dev->irq[j],
6464 					  hns_roce_v2_msix_interrupt_abn,
6465 					  0, hr_dev->irq_names[j], hr_dev);
6466 
6467 		else if (j < (other_num + comp_num))
6468 			ret = request_irq(eq_table->eq[j - other_num].irq,
6469 					  hns_roce_v2_msix_interrupt_eq,
6470 					  0, hr_dev->irq_names[j + aeq_num],
6471 					  &eq_table->eq[j - other_num]);
6472 		else
6473 			ret = request_irq(eq_table->eq[j - other_num].irq,
6474 					  hns_roce_v2_msix_interrupt_eq,
6475 					  0, hr_dev->irq_names[j - comp_num],
6476 					  &eq_table->eq[j - other_num]);
6477 		if (ret) {
6478 			dev_err(hr_dev->dev, "request irq error!\n");
6479 			goto err_request_failed;
6480 		}
6481 	}
6482 
6483 	return 0;
6484 
6485 err_request_failed:
6486 	for (j -= 1; j >= 0; j--)
6487 		if (j < other_num)
6488 			free_irq(hr_dev->irq[j], hr_dev);
6489 		else
6490 			free_irq(eq_table->eq[j - other_num].irq,
6491 				 &eq_table->eq[j - other_num]);
6492 
6493 err_kzalloc_failed:
6494 	for (i -= 1; i >= 0; i--)
6495 		kfree(hr_dev->irq_names[i]);
6496 
6497 	return ret;
6498 }
6499 
6500 static void __hns_roce_free_irq(struct hns_roce_dev *hr_dev)
6501 {
6502 	int irq_num;
6503 	int eq_num;
6504 	int i;
6505 
6506 	eq_num = hr_dev->caps.num_comp_vectors + hr_dev->caps.num_aeq_vectors;
6507 	irq_num = eq_num + hr_dev->caps.num_other_vectors;
6508 
6509 	for (i = 0; i < hr_dev->caps.num_other_vectors; i++)
6510 		free_irq(hr_dev->irq[i], hr_dev);
6511 
6512 	for (i = 0; i < eq_num; i++)
6513 		free_irq(hr_dev->eq_table.eq[i].irq, &hr_dev->eq_table.eq[i]);
6514 
6515 	for (i = 0; i < irq_num; i++)
6516 		kfree(hr_dev->irq_names[i]);
6517 }
6518 
6519 static int hns_roce_v2_init_eq_table(struct hns_roce_dev *hr_dev)
6520 {
6521 	struct hns_roce_eq_table *eq_table = &hr_dev->eq_table;
6522 	struct device *dev = hr_dev->dev;
6523 	struct hns_roce_eq *eq;
6524 	int other_num;
6525 	int comp_num;
6526 	int aeq_num;
6527 	int irq_num;
6528 	int eq_num;
6529 	u8 eq_cmd;
6530 	int ret;
6531 	int i;
6532 
6533 	other_num = hr_dev->caps.num_other_vectors;
6534 	comp_num = hr_dev->caps.num_comp_vectors;
6535 	aeq_num = hr_dev->caps.num_aeq_vectors;
6536 
6537 	eq_num = comp_num + aeq_num;
6538 	irq_num = eq_num + other_num;
6539 
6540 	eq_table->eq = kcalloc(eq_num, sizeof(*eq_table->eq), GFP_KERNEL);
6541 	if (!eq_table->eq)
6542 		return -ENOMEM;
6543 
6544 	/* create eq */
6545 	for (i = 0; i < eq_num; i++) {
6546 		eq = &eq_table->eq[i];
6547 		eq->hr_dev = hr_dev;
6548 		eq->eqn = i;
6549 		if (i < comp_num) {
6550 			/* CEQ */
6551 			eq_cmd = HNS_ROCE_CMD_CREATE_CEQC;
6552 			eq->type_flag = HNS_ROCE_CEQ;
6553 			eq->entries = hr_dev->caps.ceqe_depth;
6554 			eq->eqe_size = hr_dev->caps.ceqe_size;
6555 			eq->irq = hr_dev->irq[i + other_num + aeq_num];
6556 			eq->eq_max_cnt = HNS_ROCE_CEQ_DEFAULT_BURST_NUM;
6557 			eq->eq_period = HNS_ROCE_CEQ_DEFAULT_INTERVAL;
6558 		} else {
6559 			/* AEQ */
6560 			eq_cmd = HNS_ROCE_CMD_CREATE_AEQC;
6561 			eq->type_flag = HNS_ROCE_AEQ;
6562 			eq->entries = hr_dev->caps.aeqe_depth;
6563 			eq->eqe_size = hr_dev->caps.aeqe_size;
6564 			eq->irq = hr_dev->irq[i - comp_num + other_num];
6565 			eq->eq_max_cnt = HNS_ROCE_AEQ_DEFAULT_BURST_NUM;
6566 			eq->eq_period = HNS_ROCE_AEQ_DEFAULT_INTERVAL;
6567 		}
6568 
6569 		ret = hns_roce_v2_create_eq(hr_dev, eq, eq_cmd);
6570 		if (ret) {
6571 			dev_err(dev, "failed to create eq.\n");
6572 			goto err_create_eq_fail;
6573 		}
6574 	}
6575 
6576 	INIT_WORK(&hr_dev->ecc_work, fmea_ram_ecc_work);
6577 
6578 	hr_dev->irq_workq = alloc_ordered_workqueue("hns_roce_irq_workq", 0);
6579 	if (!hr_dev->irq_workq) {
6580 		dev_err(dev, "failed to create irq workqueue.\n");
6581 		ret = -ENOMEM;
6582 		goto err_create_eq_fail;
6583 	}
6584 
6585 	ret = __hns_roce_request_irq(hr_dev, irq_num, comp_num, aeq_num,
6586 				     other_num);
6587 	if (ret) {
6588 		dev_err(dev, "failed to request irq.\n");
6589 		goto err_request_irq_fail;
6590 	}
6591 
6592 	/* enable irq */
6593 	hns_roce_v2_int_mask_enable(hr_dev, eq_num, EQ_ENABLE);
6594 
6595 	return 0;
6596 
6597 err_request_irq_fail:
6598 	destroy_workqueue(hr_dev->irq_workq);
6599 
6600 err_create_eq_fail:
6601 	for (i -= 1; i >= 0; i--)
6602 		hns_roce_v2_destroy_eqc(hr_dev, &eq_table->eq[i]);
6603 	kfree(eq_table->eq);
6604 
6605 	return ret;
6606 }
6607 
6608 static void hns_roce_v2_cleanup_eq_table(struct hns_roce_dev *hr_dev)
6609 {
6610 	struct hns_roce_eq_table *eq_table = &hr_dev->eq_table;
6611 	int eq_num;
6612 	int i;
6613 
6614 	eq_num = hr_dev->caps.num_comp_vectors + hr_dev->caps.num_aeq_vectors;
6615 
6616 	/* Disable irq */
6617 	hns_roce_v2_int_mask_enable(hr_dev, eq_num, EQ_DISABLE);
6618 
6619 	__hns_roce_free_irq(hr_dev);
6620 	destroy_workqueue(hr_dev->irq_workq);
6621 
6622 	for (i = 0; i < eq_num; i++)
6623 		hns_roce_v2_destroy_eqc(hr_dev, &eq_table->eq[i]);
6624 
6625 	kfree(eq_table->eq);
6626 }
6627 
6628 static const struct ib_device_ops hns_roce_v2_dev_ops = {
6629 	.destroy_qp = hns_roce_v2_destroy_qp,
6630 	.modify_cq = hns_roce_v2_modify_cq,
6631 	.poll_cq = hns_roce_v2_poll_cq,
6632 	.post_recv = hns_roce_v2_post_recv,
6633 	.post_send = hns_roce_v2_post_send,
6634 	.query_qp = hns_roce_v2_query_qp,
6635 	.req_notify_cq = hns_roce_v2_req_notify_cq,
6636 };
6637 
6638 static const struct ib_device_ops hns_roce_v2_dev_srq_ops = {
6639 	.modify_srq = hns_roce_v2_modify_srq,
6640 	.post_srq_recv = hns_roce_v2_post_srq_recv,
6641 	.query_srq = hns_roce_v2_query_srq,
6642 };
6643 
6644 static const struct hns_roce_hw hns_roce_hw_v2 = {
6645 	.cmq_init = hns_roce_v2_cmq_init,
6646 	.cmq_exit = hns_roce_v2_cmq_exit,
6647 	.hw_profile = hns_roce_v2_profile,
6648 	.hw_init = hns_roce_v2_init,
6649 	.hw_exit = hns_roce_v2_exit,
6650 	.post_mbox = v2_post_mbox,
6651 	.poll_mbox_done = v2_poll_mbox_done,
6652 	.chk_mbox_avail = v2_chk_mbox_is_avail,
6653 	.set_gid = hns_roce_v2_set_gid,
6654 	.set_mac = hns_roce_v2_set_mac,
6655 	.write_mtpt = hns_roce_v2_write_mtpt,
6656 	.rereg_write_mtpt = hns_roce_v2_rereg_write_mtpt,
6657 	.frmr_write_mtpt = hns_roce_v2_frmr_write_mtpt,
6658 	.mw_write_mtpt = hns_roce_v2_mw_write_mtpt,
6659 	.write_cqc = hns_roce_v2_write_cqc,
6660 	.set_hem = hns_roce_v2_set_hem,
6661 	.clear_hem = hns_roce_v2_clear_hem,
6662 	.modify_qp = hns_roce_v2_modify_qp,
6663 	.dereg_mr = hns_roce_v2_dereg_mr,
6664 	.qp_flow_control_init = hns_roce_v2_qp_flow_control_init,
6665 	.init_eq = hns_roce_v2_init_eq_table,
6666 	.cleanup_eq = hns_roce_v2_cleanup_eq_table,
6667 	.write_srqc = hns_roce_v2_write_srqc,
6668 	.query_cqc = hns_roce_v2_query_cqc,
6669 	.query_qpc = hns_roce_v2_query_qpc,
6670 	.query_mpt = hns_roce_v2_query_mpt,
6671 	.query_hw_counter = hns_roce_hw_v2_query_counter,
6672 	.hns_roce_dev_ops = &hns_roce_v2_dev_ops,
6673 	.hns_roce_dev_srq_ops = &hns_roce_v2_dev_srq_ops,
6674 };
6675 
6676 static const struct pci_device_id hns_roce_hw_v2_pci_tbl[] = {
6677 	{PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_25GE_RDMA), 0},
6678 	{PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_25GE_RDMA_MACSEC), 0},
6679 	{PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_50GE_RDMA), 0},
6680 	{PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_50GE_RDMA_MACSEC), 0},
6681 	{PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_100G_RDMA_MACSEC), 0},
6682 	{PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_200G_RDMA), 0},
6683 	{PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_RDMA_DCB_PFC_VF),
6684 	 HNAE3_DEV_SUPPORT_ROCE_DCB_BITS},
6685 	/* required last entry */
6686 	{0, }
6687 };
6688 
6689 MODULE_DEVICE_TABLE(pci, hns_roce_hw_v2_pci_tbl);
6690 
6691 static void hns_roce_hw_v2_get_cfg(struct hns_roce_dev *hr_dev,
6692 				  struct hnae3_handle *handle)
6693 {
6694 	struct hns_roce_v2_priv *priv = hr_dev->priv;
6695 	const struct pci_device_id *id;
6696 	int i;
6697 
6698 	hr_dev->pci_dev = handle->pdev;
6699 	id = pci_match_id(hns_roce_hw_v2_pci_tbl, hr_dev->pci_dev);
6700 	hr_dev->is_vf = id->driver_data;
6701 	hr_dev->dev = &handle->pdev->dev;
6702 	hr_dev->hw = &hns_roce_hw_v2;
6703 	hr_dev->sdb_offset = ROCEE_DB_SQ_L_0_REG;
6704 	hr_dev->odb_offset = hr_dev->sdb_offset;
6705 
6706 	/* Get info from NIC driver. */
6707 	hr_dev->reg_base = handle->rinfo.roce_io_base;
6708 	hr_dev->mem_base = handle->rinfo.roce_mem_base;
6709 	hr_dev->caps.num_ports = 1;
6710 	hr_dev->iboe.netdevs[0] = handle->rinfo.netdev;
6711 	hr_dev->iboe.phy_port[0] = 0;
6712 
6713 	addrconf_addr_eui48((u8 *)&hr_dev->ib_dev.node_guid,
6714 			    hr_dev->iboe.netdevs[0]->dev_addr);
6715 
6716 	for (i = 0; i < handle->rinfo.num_vectors; i++)
6717 		hr_dev->irq[i] = pci_irq_vector(handle->pdev,
6718 						i + handle->rinfo.base_vector);
6719 
6720 	/* cmd issue mode: 0 is poll, 1 is event */
6721 	hr_dev->cmd_mod = 1;
6722 	hr_dev->loop_idc = 0;
6723 
6724 	hr_dev->reset_cnt = handle->ae_algo->ops->ae_dev_reset_cnt(handle);
6725 	priv->handle = handle;
6726 }
6727 
6728 static int __hns_roce_hw_v2_init_instance(struct hnae3_handle *handle)
6729 {
6730 	struct hns_roce_dev *hr_dev;
6731 	int ret;
6732 
6733 	hr_dev = ib_alloc_device(hns_roce_dev, ib_dev);
6734 	if (!hr_dev)
6735 		return -ENOMEM;
6736 
6737 	hr_dev->priv = kzalloc(sizeof(struct hns_roce_v2_priv), GFP_KERNEL);
6738 	if (!hr_dev->priv) {
6739 		ret = -ENOMEM;
6740 		goto error_failed_kzalloc;
6741 	}
6742 
6743 	hns_roce_hw_v2_get_cfg(hr_dev, handle);
6744 
6745 	ret = hns_roce_init(hr_dev);
6746 	if (ret) {
6747 		dev_err(hr_dev->dev, "RoCE Engine init failed!\n");
6748 		goto error_failed_roce_init;
6749 	}
6750 
6751 	if (hr_dev->pci_dev->revision == PCI_REVISION_ID_HIP08) {
6752 		ret = free_mr_init(hr_dev);
6753 		if (ret) {
6754 			dev_err(hr_dev->dev, "failed to init free mr!\n");
6755 			goto error_failed_free_mr_init;
6756 		}
6757 	}
6758 
6759 	handle->priv = hr_dev;
6760 
6761 	return 0;
6762 
6763 error_failed_free_mr_init:
6764 	hns_roce_exit(hr_dev);
6765 
6766 error_failed_roce_init:
6767 	kfree(hr_dev->priv);
6768 
6769 error_failed_kzalloc:
6770 	ib_dealloc_device(&hr_dev->ib_dev);
6771 
6772 	return ret;
6773 }
6774 
6775 static void __hns_roce_hw_v2_uninit_instance(struct hnae3_handle *handle,
6776 					   bool reset)
6777 {
6778 	struct hns_roce_dev *hr_dev = handle->priv;
6779 
6780 	if (!hr_dev)
6781 		return;
6782 
6783 	handle->priv = NULL;
6784 
6785 	hr_dev->state = HNS_ROCE_DEVICE_STATE_UNINIT;
6786 	hns_roce_handle_device_err(hr_dev);
6787 
6788 	hns_roce_exit(hr_dev);
6789 	kfree(hr_dev->priv);
6790 	ib_dealloc_device(&hr_dev->ib_dev);
6791 }
6792 
6793 static int hns_roce_hw_v2_init_instance(struct hnae3_handle *handle)
6794 {
6795 	const struct hnae3_ae_ops *ops = handle->ae_algo->ops;
6796 	const struct pci_device_id *id;
6797 	struct device *dev = &handle->pdev->dev;
6798 	int ret;
6799 
6800 	handle->rinfo.instance_state = HNS_ROCE_STATE_INIT;
6801 
6802 	if (ops->ae_dev_resetting(handle) || ops->get_hw_reset_stat(handle)) {
6803 		handle->rinfo.instance_state = HNS_ROCE_STATE_NON_INIT;
6804 		goto reset_chk_err;
6805 	}
6806 
6807 	id = pci_match_id(hns_roce_hw_v2_pci_tbl, handle->pdev);
6808 	if (!id)
6809 		return 0;
6810 
6811 	if (id->driver_data && handle->pdev->revision == PCI_REVISION_ID_HIP08)
6812 		return 0;
6813 
6814 	ret = __hns_roce_hw_v2_init_instance(handle);
6815 	if (ret) {
6816 		handle->rinfo.instance_state = HNS_ROCE_STATE_NON_INIT;
6817 		dev_err(dev, "RoCE instance init failed! ret = %d\n", ret);
6818 		if (ops->ae_dev_resetting(handle) ||
6819 		    ops->get_hw_reset_stat(handle))
6820 			goto reset_chk_err;
6821 		else
6822 			return ret;
6823 	}
6824 
6825 	handle->rinfo.instance_state = HNS_ROCE_STATE_INITED;
6826 
6827 	return 0;
6828 
6829 reset_chk_err:
6830 	dev_err(dev, "Device is busy in resetting state.\n"
6831 		     "please retry later.\n");
6832 
6833 	return -EBUSY;
6834 }
6835 
6836 static void hns_roce_hw_v2_uninit_instance(struct hnae3_handle *handle,
6837 					   bool reset)
6838 {
6839 	if (handle->rinfo.instance_state != HNS_ROCE_STATE_INITED)
6840 		return;
6841 
6842 	handle->rinfo.instance_state = HNS_ROCE_STATE_UNINIT;
6843 
6844 	__hns_roce_hw_v2_uninit_instance(handle, reset);
6845 
6846 	handle->rinfo.instance_state = HNS_ROCE_STATE_NON_INIT;
6847 }
6848 static int hns_roce_hw_v2_reset_notify_down(struct hnae3_handle *handle)
6849 {
6850 	struct hns_roce_dev *hr_dev;
6851 
6852 	if (handle->rinfo.instance_state != HNS_ROCE_STATE_INITED) {
6853 		set_bit(HNS_ROCE_RST_DIRECT_RETURN, &handle->rinfo.state);
6854 		return 0;
6855 	}
6856 
6857 	handle->rinfo.reset_state = HNS_ROCE_STATE_RST_DOWN;
6858 	clear_bit(HNS_ROCE_RST_DIRECT_RETURN, &handle->rinfo.state);
6859 
6860 	hr_dev = handle->priv;
6861 	if (!hr_dev)
6862 		return 0;
6863 
6864 	hr_dev->active = false;
6865 	hr_dev->dis_db = true;
6866 	hr_dev->state = HNS_ROCE_DEVICE_STATE_RST_DOWN;
6867 
6868 	return 0;
6869 }
6870 
6871 static int hns_roce_hw_v2_reset_notify_init(struct hnae3_handle *handle)
6872 {
6873 	struct device *dev = &handle->pdev->dev;
6874 	int ret;
6875 
6876 	if (test_and_clear_bit(HNS_ROCE_RST_DIRECT_RETURN,
6877 			       &handle->rinfo.state)) {
6878 		handle->rinfo.reset_state = HNS_ROCE_STATE_RST_INITED;
6879 		return 0;
6880 	}
6881 
6882 	handle->rinfo.reset_state = HNS_ROCE_STATE_RST_INIT;
6883 
6884 	dev_info(&handle->pdev->dev, "In reset process RoCE client reinit.\n");
6885 	ret = __hns_roce_hw_v2_init_instance(handle);
6886 	if (ret) {
6887 		/* when reset notify type is HNAE3_INIT_CLIENT In reset notify
6888 		 * callback function, RoCE Engine reinitialize. If RoCE reinit
6889 		 * failed, we should inform NIC driver.
6890 		 */
6891 		handle->priv = NULL;
6892 		dev_err(dev, "In reset process RoCE reinit failed %d.\n", ret);
6893 	} else {
6894 		handle->rinfo.reset_state = HNS_ROCE_STATE_RST_INITED;
6895 		dev_info(dev, "reset done, RoCE client reinit finished.\n");
6896 	}
6897 
6898 	return ret;
6899 }
6900 
6901 static int hns_roce_hw_v2_reset_notify_uninit(struct hnae3_handle *handle)
6902 {
6903 	if (test_bit(HNS_ROCE_RST_DIRECT_RETURN, &handle->rinfo.state))
6904 		return 0;
6905 
6906 	handle->rinfo.reset_state = HNS_ROCE_STATE_RST_UNINIT;
6907 	dev_info(&handle->pdev->dev, "In reset process RoCE client uninit.\n");
6908 	msleep(HNS_ROCE_V2_HW_RST_UNINT_DELAY);
6909 	__hns_roce_hw_v2_uninit_instance(handle, false);
6910 
6911 	return 0;
6912 }
6913 
6914 static int hns_roce_hw_v2_reset_notify(struct hnae3_handle *handle,
6915 				       enum hnae3_reset_notify_type type)
6916 {
6917 	int ret = 0;
6918 
6919 	switch (type) {
6920 	case HNAE3_DOWN_CLIENT:
6921 		ret = hns_roce_hw_v2_reset_notify_down(handle);
6922 		break;
6923 	case HNAE3_INIT_CLIENT:
6924 		ret = hns_roce_hw_v2_reset_notify_init(handle);
6925 		break;
6926 	case HNAE3_UNINIT_CLIENT:
6927 		ret = hns_roce_hw_v2_reset_notify_uninit(handle);
6928 		break;
6929 	default:
6930 		break;
6931 	}
6932 
6933 	return ret;
6934 }
6935 
6936 static const struct hnae3_client_ops hns_roce_hw_v2_ops = {
6937 	.init_instance = hns_roce_hw_v2_init_instance,
6938 	.uninit_instance = hns_roce_hw_v2_uninit_instance,
6939 	.reset_notify = hns_roce_hw_v2_reset_notify,
6940 };
6941 
6942 static struct hnae3_client hns_roce_hw_v2_client = {
6943 	.name = "hns_roce_hw_v2",
6944 	.type = HNAE3_CLIENT_ROCE,
6945 	.ops = &hns_roce_hw_v2_ops,
6946 };
6947 
6948 static int __init hns_roce_hw_v2_init(void)
6949 {
6950 	return hnae3_register_client(&hns_roce_hw_v2_client);
6951 }
6952 
6953 static void __exit hns_roce_hw_v2_exit(void)
6954 {
6955 	hnae3_unregister_client(&hns_roce_hw_v2_client);
6956 }
6957 
6958 module_init(hns_roce_hw_v2_init);
6959 module_exit(hns_roce_hw_v2_exit);
6960 
6961 MODULE_LICENSE("Dual BSD/GPL");
6962 MODULE_AUTHOR("Wei Hu <xavier.huwei@huawei.com>");
6963 MODULE_AUTHOR("Lijun Ou <oulijun@huawei.com>");
6964 MODULE_AUTHOR("Shaobo Xu <xushaobo2@huawei.com>");
6965 MODULE_DESCRIPTION("Hisilicon Hip08 Family RoCE Driver");
6966